This is the first of a series of articles exploring the benefits of Edge AI for a variety of applications.&nbsp;<style type="text/css" id="isPasted">
p.p1 {margin: 0.0px 0.0px 0.0px 0.0px; text-align: justify; font: 11.0px Arial; color: #000000} p.p2 {margin: 0.0px 0.0px 0.0px 0.0px; text-align: justify; font: 11.0px Arial; color: #000000; min-height: 12.0px} p.p3 {margin: 0.0px 0.0px 0.0px 0.0px; font: 11.0px Arial; color: #000000; min-height: 12.0px} p.p4 {margin: 0.0px 0.0px 0.0px 0.0px; font: 12.0px Arial; color: #000000} p.p5 {margin: 0.0px 0.0px 0.0px 0.0px; font: 14.0px Arial; color: #000000; min-height: 16.0px} p.p6 {margin: 0.0px 0.0px 0.0px 0.0px; text-align: justify; font: 12.0px Arial; color: #000000} p.p7 {margin: 0.0px 0.0px 18.0px 0.0px; text-align: justify; font: 11.0px Arial; color: #000000} p.p8 {margin: 0.0px 0.0px 6.0px 0.0px; font: 12.0px Arial; color: #000000} li.li1 {margin: 0.0px 0.0px 0.0px 0.0px; text-align: justify; font: 11.0px Arial; color: #000000} span.s1 {text-decoration: underline ; color: #0000ff} span.s2 {font: 11.0px Symbol} ul.ul1 {list-style-type: disc}
</style>The proliferation of digital transactions, smart phones, cyber-physical systems, and other internet connected objects in modern enterprises has led to generation of unprecedented volumes of data. Nowadays, companies are increasingly taking advantage of this data to improve the efficacy of their business processes and make better decisions via analytics. Machine learning (ML) and Artificial Intelligence (AI) algorithms play an instrumental role in collecting and analyzing enterprise data in fast, automated, and cost-efficient ways.&nbsp;Most enterprise-scale ML/AI systems are deployed in conjunction with cloud computing infrastructures. This is because cloud computing eases the storage and management of large datasets, as well as the processing of many data points. Nevertheless, cloud-based AI systems are not the best choice when it comes to processing data close to the field and providing real-time performance. Likewise, cloud systems are vulnerable to significant cyber-security threats as large volumes of data are transferred from different field systems to a single cloud infrastructure.&nbsp;The edge computing paradigm alleviates the latency and security limitations of cloud computing. Edge computing brings data collection and decision-making closer to the field where data are produced. In this way, it also moves computing power close to the data sources, which reduces latency. Moreover, it limits the amount of data that are transferred from the field to the cloud, which foundational for reducing data breaches and stronger data protection.&nbsp;Reduced latency and increased security is the reason behind the ongoing shift of AI/ML applications <a href="https://www.idc.com/getdoc.jsp?containerId=prUS49532322" target="_blank">from the cloud to the edge</a>. In recent years edge AI deployments are gaining traction across various verticals. For instance, connected and autonomous vehicles leverage edge AI to reduce latency as split-second decision-making can make a huge impact on passenger safety. Similarly, across the manufacturing sectors, companies are taking advantage of ML algorithms on the edge to provide near real-time detection of machine failures and of defective products.&nbsp;<h2>The Security Benefits of Edge AI</h2>In general, state-of-the-art AI/ML applications needs strong security for a variety of reasons, including:<ul><li>Their data intensive nature, which makes them susceptible to costly breaches. This can be critical especially in applications that handle sensitive information such as healthcare and personal data management.&nbsp;</li><li>Machine learning itself has security vulnerabilities due to its reliance on suitable training data for proper operation. For example, in instances known as poisoning attacks adversaries compromise the operation of an ML system through by its training data with malicious datasets. In another example, evasion attacks confuse ML/AI systems by providing them with adversarial examples e.g., perturbed malicious input that looks like an untampered copy, yet it side-tracks the operation of the ML algorithms. &nbsp;&nbsp;</li></ul>Edge computing is recognized for its ability to provide stronger security and data protection than conventional cloud deployments. Specifically, edge AI provides the following security benefits over cloud AI deployments:<ul><li>Operation based on safer (local) data. In edge AI data need not travel over a wide area network, which reduces the possibility of cybersecurity attacks. AI algorithms operate based on local data, which reside within edge servers or on the device itself. This makes them much more difficult to tamper with than data that are transferred across networks.</li><li>Almost impossible for malicious party to access all data. Edge AI datasets are distributed across various on-premises data centers or devices, which means that there is no single point of vulnerability. Edge AI systems use distributed data that reside at the edge of the network, which makes it impossible for an adversary to gain access to all the data by attacking one point of the network. &nbsp;</li><li>Much fewer breach points. Edge AI deployments limit drastically the number of connections and the number of data transfers from edge to the cloud. This means that they provide much fewer breach points than conventional cloud AI deployments. Moreover, the fewer number of connections facilitates the deployment of a set of carefully encrypted connections between the edge devices and the cloud data centers. In this way, the vulnerable attack surface of edge AI applications is significantly reduced.</li><li>Flexible and cost-effective regulatory compliance. Edge AI facilitates enterprises in their efforts to comply with mandatory privacy and data protection regulations such as the GDPR (General Data Protection Regulation) in Europe. In particular, edge computing enables end-users and data providers to control their personal data rather than relying on the services of a vendor or infrastructure provider. Moreover, it enables the implementation of decentralized security policies that tend to be more resilient. In this way, the risk of non-compliance for AI/ML services providers is significantly reduced.&nbsp;</li></ul><h2>Increasing Resilience with Different Edge AI Deployment Paradigms&nbsp;</h2>Edge AI applications are usually deployed in conjunction with cloud infrastructures, as some algorithms run in the cloud, where data points from various distributed sources and edge computing processes are aggregated. Moreover, edge AI applications come at varying deployment configurations based on the placement of ML/AI processes. For instance, AI analytics functions can be run inside a microsystem (e.g., a sensor), within an edge computing cluster or gateway, or even in the cloud. &nbsp;Different ML paradigms enable a wide set of edge AI deployment configurations. These ML paradigms include:&nbsp;<ul><li>Embedded Machine Learning, which means executing ML algorithms within embedded devices. This paradigm enables data processing and data analytics within devices embedded with powerful processors ranging from CPUs and GPUs to neural native AI processing chips. Set-top boxes, smartphones, connected cars OBUs (On-Board Units), and many other types of fog nodes and internet-connected devices make use of this paradigm.&nbsp;</li><li>TinyML, which refers to the execution of full stack machine learning systems (e.g., deep neural networks) within very small processors such as microcontrollers and AI Accelerators amongst others. In most cases, TinyML systems serve applications that do not require complex computations. TinyML delivers the security benefits of edge AI at the highest possible degree. This is because data are virtually processed inside the data source and there is absolutely no need for data I/O (input/output) operations.</li><li>Federated Machine Learning, is a novel ML technique that trains an algorithm across multiple decentralized edge devices. This technique is very powerful as it offers the benefits of a consolidated data lake while allowing data owners have full control of their data locally, offering great security and performance.</li></ul>In non-trivial edge AI applications, it is possible to combine more than one of the above paradigms in a solution configuration. This provides increased versatility in specifying deployment configurations that meet stringent security and data protection needs.&nbsp;<h2>Case Study: Security and Data Protection for Healthcare Applications&nbsp;</h2>The future of healthcare relies on the collection and processing of massive amounts of data from patients, including for example, data from medical and consumer devices (e.g., smart watches), clinical data, laboratory data, diagnostic devices (e.g., medical imaging), genomic databases, data about the patient&rsquo;s medical history, and many more.&nbsp;The analysis of this data though AI/ML algorithms is already driving significant improvements in many healthcare sectors, including prevention, diagnosis, prognosis, and treatment. For example, the analysis of clinical, genomic, imaging, and lifestyle data provides cardiologists with predictive insights on risk factors for various cardiovascular diseases. However, the regulatory approval and consumer adoption of AI/ML enhances services hinges on strong security and data protection measures.Many state-of-the-art ML applications in the healthcare domain rely on cloud infrastructures, which creates considerable risks for data breaches and other cyber attacks. This is because large data volumes are transferred from the patients to the cloud, which makes it easier for malicious parties to tamper with healthcare data. In this context, the deployment of ML systems and algorithms at the edge of the network is becoming a good practice that alleviates such security concerns.&nbsp;In an edge AI deployment, patient data is initially processed locally within edge devices like IoT (Internet of Things) gateways. Hence, patient data do not travel outside the perimeter of the network of the hospital or home care infrastructure. Nevertheless, select AI-based insights can be properly encrypted and transferred to the cloud infrastructure to enable further processing for statistical or clinical purposes (e.g., clinical trials). For example, if a patient&rsquo;s respiratory sounds are to be analyzed, instead of transferring the entire recording to the cloud, edge AI may be used to infer significant events: number of coughs, sneezes, wheezes, and so on. In this way, both the volume of data that needs to be transferred to the clouds is reduced, reducing bandwidth requirements, and the patient&rsquo;s data, which may include conversations is protected on the premise or within a device.&nbsp;Healthcare data processing infrastructures must be compliant to applicable regulations like HIIPA (Health Insurance Portability and Accountability) in US and GDPR in Europe. This can be quite challenging given that patient information comprises personally identifiable data. Fortunately, Edge AI can help reduce the risks of data tampering and boosts regulatory compliance. Most importantly, patients and healthcare providers can choose when to share their data and for what purpose. This eases the implementation of informed consent processes that are fundamental for compliance to data protection regulations.&nbsp;<h2>Conclusion&nbsp;</h2>Edge computing can provide stronger security and data protection than conventional cloud deployments making it the right choice for security-sensitive applications such as Ip protection, healthcare, and many more. This is the first in a series of articles exploring the benefits of Edge AI across applications, industries, and products.&nbsp;The first article discussed <a href="https://www.wevolver.com/article/reducing-ais-vulnerable-attack-surface-with-edge-computing" rel="noopener noreferrer" target="_blank">about r</a><a href="https://www.wevolver.com/article/reducing-ais-vulnerable-attack-surface-with-edge-computing" rel="noopener noreferrer" target="_blank">educing AI&#39;s Vulnerable Attack Surface with Edge Computing.</a>The second article talked about <a href="https://www.wevolver.com/article/why-edge-ai-for-wearables" rel="noopener noreferrer" target="_blank">Edge AI in wearables.</a>The third article explored <a href="https://www.wevolver.com/article/how-edge-ai-is-enabling-cutting-edge-advances-in-sustainability" rel="noopener noreferrer" target="_blank">h</a><a href="https://www.wevolver.com/article/how-edge-ai-is-enabling-cutting-edge-advances-in-sustainability" rel="noopener noreferrer" target="_blank">ow edge AI is enabling cutting-edge advances in sustainability.</a>The forth article explained <a href="https://www.wevolver.com/article/why-edge-ai-is-a-win-for-automotive" rel="noopener noreferrer" target="_blank">why Edge AI is a win for automotive.</a>The fifth article analyzed <a href="https://www.wevolver.com/article/computer-vision-on-compute-constrained-embedded-devices-high-performance-and-increased-security-at-a-lower-cost" rel="noopener noreferrer" target="_blank">c</a><a href="https://www.wevolver.com/article/computer-vision-on-compute-constrained-embedded-devices-high-performance-and-increased-security-at-a-lower-cost" rel="noopener noreferrer" target="_blank">omputer Vision on Compute-Constrained Embedded Devices.</a>The sixth article explained <a href="https://www.wevolver.com/article/why-edge-ai-for-ev-battery-management" rel="noopener noreferrer" target="_blank">why e</a><a href="https://www.wevolver.com/article/why-edge-ai-for-ev-battery-management" rel="noopener noreferrer" target="_blank">dge AI is essential for EV Battery Management.</a><h2>Security with Syntiant</h2>Syntiant combines advanced silicon solutions and deep learning models to provide ultra-low-power, high performance, deep neural network processing for edge AI applications across a wide range of consumer and industrial use cases, from earbuds to automobiles. The company&rsquo;s Neural Decision Processors (NDP) are optimally designed to deploy deep (machine) learning models on the edge, where power and area are often constrained. Syntiant&rsquo;s NDP solutions can equip, every device, from earbuds and doorbells to automobiles and healthcare wearables, with powerful deep learning capabilities, enabling real-time data processing and decision making, with near zero latency, enabling secure and private artificial intelligent solutions.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjY0NzMzNDk2NDc3LVVudGl0bGVkIGRlc2lnbi02ICgxKS5wbmciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 300px;" class="fr-fic fr-dib">&nbsp;

Article #1 of the "Why Edge?" Series. Edge AI presents a multitude of benefits for security across industry. 


Reducing AI's Vulnerable Attack Surface with Edge Computing

<h3 id="isPasted">SUMMARY</h3>Illegal dumping, which is called fly tipping in Britain, was a real problem for five London neighborhoods. That&rsquo;s why the South London Partnership is currently funding and rolling out a significant &lsquo;Internet of Things&rsquo; (IoT) InnOVaTe program to manage and mitigate new challenges within the area. To combat illegal dumping they enlisted the aid of UK IoT surveillance company iDefigo, who used Matroid&rsquo;s Computer Vision software as their AI platform to produce a 21st century dumping detector.<h3> </h3><iframe width="640" height="360" src="https://www.youtube.com/embed/NhXazQMtlb0??controls=0&mute=1&autoplay=1&showinfo=0&loop=1&playlist=NhXazQMtlb0&wmode=opaque&rel=0" frameborder="0" allowfullscreen="" class="fr-draggable" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true"></iframe> <h3 id="isPasted">CHALLENGE</h3>Recent government figures show fly tipping costs London taxpayers about $460 million a year - with over 60,000 illegal dumping incidents involving construction, demolition, and excavation material in 2020 and 2021, an increase of 18% from 2019-2020.Items such as soil and aggregates, mattresses, chemicals, discarded house contents, furniture, and miscellaneous builder waste are dumped in streets, down alleyways and on housing estates. This is unsightly, and often a danger to local residents. &nbsp;Professional gangs dump tons of waste on industrial and remote sites as well.As one of the zero tolerance programs, the local councils promote a campaign called &lsquo;Let&rsquo;s SCRAP fly tipping&rsquo; designed to alert residents and business owners to the different forms of fly tipping, and to encourage them to report it. But better surveillance was needed.<h3> </h3><h3 id="isPasted">SOLUTION</h3>To supplement the &lsquo;Let&#39;s SCRAP fly tipping&rsquo; initiatives, employing surveillance cameras with the ability of recording, policing, and combating illegal dumping is an important part of the overall solution.iDefigo requested Matroid&rsquo;s assistance to quickly build an illegal dumping detector. A proof of concept model with the South London Partnership launched in February and ran through May of this year. Since then, the detectors have been in full production.&ldquo;Using Matroid&#39;s end-to-end computer vision studio, we were able to build and deploy a fly tipping model for iDefigo in record time,&rdquo; said Matroid&rsquo;s Jeff Zeller.The smart cameras can detect, inform, and support prosecutions or fines in hot spots. This enables local enforcement officers to monitor activity and help issue penalties, which reduces illegal dumping and provides a substantial return on the investment of IoT technology. <h3 id="isPasted">RESULT</h3>Matroid and iDefigo&rsquo;s AI-powered Smart Camera solution has delivered immediate benefits for the South London Partnership where the cameras have been positioned. There has been a significant reduction in illegal dumping incidents in both residential and industrial estate areas.Some areas have seen an 80 percent reduction in illegal dumping, eliminating the need for costly and time consuming clean-ups and waste disposal.&nbsp;The BBC recently filmed locally about the project and captured the sentiments of locals.&ldquo;What a difference it&rsquo;s made,&rdquo; said one resident. &ldquo;Instead of waking up each morning to find all this &lsquo;stuff&rsquo; that appeared overnight.&rdquo;The cameras have also helped trace builders who were dumping asbestos in housing areas where they weren&rsquo;t even building because they knew they couldn&rsquo;t get away with placing the asbestos at their own site.&ldquo;Matroid enabled iDefigo to quickly and efficiently build and deploy a custom computer vision model to detect fly tipping incidents across a group of London Boroughs,&quot; said iDefigo Group Commercial Director Scott Wattie. &quot;The &nbsp;technologies were a key driver in our local government client decreasing fly tipping events by 80% in areas monitored by iDefigo&rsquo;s Smart Camera solution.&quot;

Matroid Computer Vision | SEE IT.  DETECT IT.

Computer vision brings an 80% reduction in illegal dumping.  IoT surveillance company iDefigo partners with Matroid to apply AI-based computer vision for illegal trash dumping detection.  

AI Takes on Illegal Trash Dumping

This article was first published on&nbsp;<a href="https://news.mit.edu/2022/home-wireless-parkinsons-progression-0921" id="isPasted" rel="noopener noreferrer" target="_blank">MIT News</a>. &nbsp;<hr>Parkinson&rsquo;s disease is the fastest-growing neurological disease, now affecting more than 10 million people worldwide, yet clinicians still face huge challenges in tracking its severity and progression.Clinicians typically evaluate patients by testing their motor skills and cognitive functions during clinic visits. These semisubjective measurements are often skewed by outside factors &mdash; perhaps a patient is tired after a long drive to the hospital. More than <a href="https://pubmed.ncbi.nlm.nih.gov/21832214/" target="_blank">40 percent</a> of individuals with Parkinson&rsquo;s are never treated by a neurologist or Parkinson&rsquo;s specialist, often because they live too far from an urban center or have difficulty traveling.In an effort to address these problems, researchers from MIT and elsewhere demonstrated an in-home device that can monitor a patient&rsquo;s movement and gait speed, which can be used to evaluate Parkinson&rsquo;s severity, the progression of the disease, and the patient&rsquo;s response to medication.The device, which is about the size of a Wi-Fi router, gathers data passively using radio signals that reflect off the patient&rsquo;s body as they move around their home. The patient does not need to wear a gadget or change their behavior. (A <a href="https://news.mit.edu/2022/artificial-intelligence-can-detect-parkinsons-from-breathing-patterns-0822" target="_blank">recent study</a>, for example, showed that this type of device could be used to detect Parkinson&rsquo;s from a person&rsquo;s breathing patterns while sleeping.)The researchers used these devices to conduct a one-year <a href="https://www.science.org/doi/10.1126/scitranslmed.adc9669" target="_blank">at-home study</a> with 50 participants. They showed that, by using machine-learning algorithms to analyze the troves of data they passively gathered (more than 200,000 gait speed measurements), a clinician could track Parkinson&rsquo;s progression and medication response more effectively than they would with periodic, in-clinic evaluations.&ldquo;By being able to have a device in the home that can monitor a patient and tell the doctor remotely about the progression of the disease, and the patient&rsquo;s medication response so they can attend to the patient even if the patient can&rsquo;t come to the clinic &mdash; now they have real, reliable information &mdash; that actually goes a long way toward improving equity and access,&rdquo; says senior author Dina Katabi, the Thuan and Nicole Pham Professor in the Department of Electrical Engineering and Computer Science (EECS), and a principle investigator in the Computer Science and Artificial Intelligence Laboratory (CSAIL) and the MIT Jameel Clinic.The co-lead authors are EECS graduate students Yingcheng Liu and Guo Zhang. The research is published today in Science Translational Medicine.<iframe src="https://www.youtube.com/embed/8lFzABIEGBE?&wmode=opaque&rel=0&enablejsapi=1&origin=https://www.wevolver.com" frameborder="0" allowfullscreen="" class="fr-draggable" data-gtm-yt-inspected-8="true" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true" style="width: 766px; height: 435px;"></iframe><h2 id="isPasted">A human radar</h2>This work utilizes a wireless device previously developed in the Katabi lab that analyzes radio signals that bounce off people&rsquo;s bodies. It transmits signals that use a tiny fraction of the power of a Wi-Fi router &mdash; these super-low-power signals don&rsquo;t interfere with other wireless devices in the home. While radio signals pass through walls and other solid objects, they are reflected off humans due to the water in our bodies. &nbsp;This creates a &ldquo;human radar&rdquo; that can track the movement of a person in a room. Radio waves always travel at the same speed, so the length of time it takes the signals to reflect back to the device indicates how the person is moving.The device incorporates a machine-learning classifier that can pick out the precise radio signals reflected off the patient even when there are other people moving around the room. Advanced algorithms use these movement data to compute gait speed &mdash; how fast the person is walking.Because the device operates in the background and runs all day, every day, it can collect a massive amount of data. The researchers wanted to see if they could apply machine learning to these datasets to gain insights about the disease over time.They gathered 50 participants, 34 of whom had Parkinson&rsquo;s, and conducted a one-year study of in-home gait measurements Through the study, the researchers collected more than 200,000 individual measurements that they averaged to smooth out variability due to the conditions irrelevant to the disease. (For example, a patient may hurry up to answer an alarm or walk slower when talking on the phone.)They used statistical methods to analyze the data and found that in-home gait speed can be used to effectively track Parkinson&rsquo;s progression and severity. For instance, they showed that gait speed declined almost twice as fast for individuals with Parkinson&rsquo;s, compared to those without.&nbsp;&ldquo;Monitoring the patient continuously as they move around the room enabled us to get really good measurements of their gait speed. And with so much data, we were able to perform aggregation that allowed us to see very small differences,&rdquo; Zhang says.<h2>Better, faster results</h2>Drilling down on these variabilities offered some key insights. For instance, the researchers showed that daily fluctuations in a patient&rsquo;s walking speed correspond with how they are responding to their medication &mdash; walking speed may improve after a dose and then begin to decline after a few hours, as the medication impact wears off.&ldquo;This enables us to objectively measure how your mobility responds to your medication. Previously, this was very cumbersome to do because this medication effect could only be measured by having the patient keep a journal,&rdquo; Liu says.A clinician could use these data to adjust medication dosage more effectively and accurately. This is especially important since drugs used to treat disease symptoms can cause serious side effects if the patient receives too much.The researchers were able to demonstrate statistically significant results regarding Parkinson&rsquo;s progression after studying 50 people for just one year. By contrast, an often-cited study by the Michael J. Fox Foundation involved more than 500 individuals and monitored them for more than five years, Katabi says.&ldquo;For a pharmaceutical company or a biotech company trying to develop medicines for this disease, this could greatly reduce the burden and cost and speed up the development of new therapies,&rdquo; she adds.Katabi credits much of the study&rsquo;s success to the dedicated team of scientists and clinicians who worked together to tackle the many difficulties that arose along the way. For one, they began the study before the Covid-19 pandemic, so team members initially visited people&rsquo;s homes to set up the devices. When that was no longer possible, they developed a user-friendly phone app to remotely help participants as they deployed the device at home.Through the course of the study, they learned to automate processes and reduce effort, especially for the participants and clinical team.This knowledge will prove useful as they look to deploy devices in at-home studies of other neurological disorders, such as Alzheimer&rsquo;s, ALS, and Huntington&rsquo;s. They also want to explore how these methods could be used, in conjunction with other <a href="https://news.mit.edu/2022/artificial-intelligence-can-detect-parkinsons-from-breathing-patterns-0822" target="_blank">work from the Katabi lab</a> showing that Parkinson&rsquo;s can be diagnosed by monitoring breathing, to collect a holistic set of markers that could diagnose the disease early and then be used to track and treat it.&ldquo;This radio-wave sensor can enable more care (and research) to migrate from hospitals to the home where it is most desired and needed,&rdquo; says Ray Dorsey, a professor of neurology at the University of Rochester Medical Center, co-author of Ending Parkinson&rsquo;s, and a co-author of this research paper. &ldquo;Its potential is just beginning to be seen. We are moving toward a day where we can diagnose and predict disease at home. In the future, we may even be able to predict and ideally prevent events like falls and heart attacks.&rdquo;<hr>&quot;Reprinted with permission of&nbsp;<a href="https://news.mit.edu/2022/home-wireless-parkinsons-progression-0921" id="isPasted" rel="noopener noreferrer" target="_blank">MIT News</a>&quot;&nbsp;

By continuously monitoring a patient’s gait speed, an in-home wireless system can assess the condition’s severity between visits to the doctor’s office. Image: N.Fuller, SayoStudio

By continuously monitoring a patient’s gait speed, the system can assess the condition’s severity between visits to the doctor’s office.

In-home wireless device tracks disease progression in Parkinson's patients

Researchers from the McKelvey School of Engineering at Washington University in St. Louis have developed a machine learning algorithm that can create a continuous 3D model of cells from a partial set of 2D images that were taken using the same standard microscopy tools found in many labs today.Their findings were published Sept. 16 in the journal <a href="https://www.nature.com/articles/s42256-022-00530-3" target="_blank">Nature Machine Intelligence</a>.&ldquo;We train the model on the set of digital images to obtain a continuous representation,&rdquo; said <a href="https://engineering.wustl.edu/faculty/Ulugbek-Kamilov.html" target="_blank">Ulugbek Kamilov</a>, assistant professor of electrical and systems engineering and of computer science and engineering. &ldquo;Now, I can show it any way I want. I can zoom in smoothly and there is no pixelation.&rdquo;The key to this work was the use of a neural field network, a particular kind of machine learning system that learns a mapping from spatial coordinates to the corresponding physical quantities. When the training is complete, researchers can point to any coordinate and the model can provide the image value at that location.A particular strength of neural field networks is that they do not need to be trained on copious amounts of similar data. Instead, as long as there is a sufficient number of 2D images of the sample, the network can represent it in its entirety, inside and out.The image used to train the network is just like any other microscopy image. In essence, a cell is lit from below; the light travels through it and is captured on the other side, creating an image.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1663764715370-2Dto3DKamilovcrop.jpg" style="width: 766px;" class="fr-fic fr-dib">The imaging system can zoom in on a pixelated image and fill in the missing pieces, creating a continuous 3D representation.&nbsp;&ldquo;Because I have some views of the cell, I can use those images to train the model,&rdquo; Kamilov said. This is done by feeding the model information about a point in the sample where the image captured some of the internal structure of the cell.Then the network takes its best shot at recreating that structure. If the output is wrong, the network is tweaked. If it&rsquo;s correct, that pathway is reinforced. Once the predictions match real-world measurements, the network is ready to fill in parts of the cell that were not captured by the original 2D images.The model now contains information of a full, continuous representation of the cell &mdash; there&rsquo;s no need to save a data-heavy image file because it can always be recreated by the neural field network.And, Kamilov said, not only is the model an easy-to-store, true representation of the cell, but also, in many ways, it&rsquo;s more useful than the real thing.&ldquo;I can put any coordinate in and generate that view,&rdquo; he said. &ldquo;Or I can generate entirely new views from different angles.&rdquo; He can use the model to spin a cell like a top or zoom in for a closer look; use the model to do other numerical tasks; or even feed it into another algorithm.

The imaging system can zoom in on a pixelated image and fill in the missing pieces, creating a continuous 3D representation.

Neural field network doesn’t need to be trained on other samples

Machine learning generates 3D model from 2D pictures

A short guide to understanding the edge neural network design. 


Neural Network Architectures at the Edge: Modelling for Energy Efficiency and Machine Learning Performance

This article is based on two blogs by Rouzbeh Shirvani and Sean McGregor, with additions and editing by John Soldatos.&nbsp;&nbsp;<style type="text/css" id="isPasted">
p.p1 {margin: 0.0px 0.0px 0.0px 0.0px; text-align: justify; font: 11.0px Arial; color: #000000} p.p2 {margin: 0.0px 0.0px 0.0px 0.0px; text-align: justify; font: 11.0px Arial; color: #000000; min-height: 12.0px} p.p3 {margin: 0.0px 0.0px 0.0px 0.0px; font: 14.0px Arial; color: #000000; min-height: 16.0px} p.p4 {margin: 0.0px 0.0px 0.0px 0.0px; font: 12.0px Arial; color: #000000; min-height: 14.0px} p.p5 {margin: 0.0px 0.0px 0.0px 0.0px; font: 12.0px Arial; color: #000000} p.p6 {margin: 0.0px 0.0px 0.0px 0.0px; font: 11.0px Arial; color: #000000; min-height: 12.0px} li.li1 {margin: 0.0px 0.0px 0.0px 0.0px; text-align: justify; font: 11.0px Arial; color: #000000} span.s1 {font: 11.0px Symbol} span.s2 {text-decoration: underline ; color: #0000ff} span.s3 {text-decoration: underline} span.s4 {font: 7.3px Arial} ul.ul1 {list-style-type: disc}
</style>In recent years, there is a surge of interest in Machine Learning (ML) and Artificial Intelligence (AI) solutions. This interest is driven by advances in parallel hardware, scalable software systems, advanced machine learning frameworks, as well as cloud/edge computing. Furthermore, the explosion of digital data collected makes it possible to develop more effective and accurate ML systems based on deep neural networks. This is because deep learning (DL) outperforms classical ML when large volumes of training data are available.&nbsp;State-of-the-art ML/DL systems span both cloud computing and edge computing infrastructures, and can have varying characteristics, such as ML model performance (e.g., classification accuracy), training speed, and energy efficiency. These characteristics determine whether a system can meet specific application requirements such as latency and environmental performance. In most cases it is not possible to optimize all these parameters simultaneously. Therefore ML/DL developers often have to determine the relevant trade-offs and optimize the performance according to their specific application needs.&nbsp;The choice of processors is key in setting the performance characteristics of a model. Larger, more power-hungry processors may appear to be the higher performing option, but specifically designed neural processors, or accelerators, often provide the better power/performance tradeoff. Complicating matters, deep learning systems are statistical and non-deterministic, which means that there is rarely a single optimal system configuration that yields the best performance and accuracy. Rather, it is possible to deploy alternative configurations of hardware and algorithms with comparable performance and efficiency.<h2>Introducing ML/DL and Neural Architectures Benchmarks</h2>The variety of available neural network architectures, configurations, and performance-related parameters makes it challenging to compare alternative ML/DL products and services. To facilitate this, engineers often turn to proven benchmarks for compute intense and data intense systems. For example:<ul><li>The <a href="http://www.spec.org/spec/faq/#01SPEC.General.10whatis" target="_blank">Standard Performance Evaluation Corporation (SPEC</a>) family of benchmarks focuses on computational workloads over different computing architectures including cloud-based systems.</li><li>The <a href="https://netlib.org/utk/people/JackDongarra/faq-linpack.html" target="_blank">LINPACK Benchmarks</a> provide measures of a computer&rsquo;s floating-point rate of execution i.e., they focus on floating-point computing power.&nbsp;</li><li>The <a href="https://www.tpc.org/" target="_blank">TPC consortium</a> includes industry leaders in computing systems and datasets, which have provided a suite of benchmarks for stress testing the transaction processing and data warehousing capabilities of data intensive systems.</li><li>The <a href="https://hpcg-benchmark.org/" target="_blank">High Performance Conjugate Gradients (HPCG)</a> benchmark provides novel metrics that enable the evaluation and ranking of High Performance Computing (HPC) systems.</li></ul>These benchmarks aid in evaluating different components of ML/DL systems, which are typically deployed over computationally intensive platforms. However, they are overly focused on conventional computational workloads and hence not sufficient for training and executing deep learning tasks that are typically more complex. &nbsp;To provide proper benchmarks for ML systems an end-to-end approach is needed. It must consider all the different steps that are involved in the development and execution of an ML/DL pipeline i.e., from the training of an accurate model to the execution of different implementation and computation options towards optimizing system performance (e.g., execution speed).&nbsp;Independent benchmarks provide a credible and impartial evaluation of alternative solutions, providing stakeholders with invaluable insights on the available market offerings. Moreover, they provide guidance for the co-design of ML/DL solutions by ML modelers, data scientists and embedded systems engineers.&nbsp;Specifically, the benchmarks provide these teams with tangible metrics that can help them evaluate alternative system designs for a variety of popular use cases. Furthermore, ML benchmarking suites enable market players to compete against the same set of metrics, fostering innovation and healthy competition.&nbsp;<h2>The MLPerf Benchmarking Suite</h2><a href="https://mlcommons.org/en/" target="_blank">MLPerf is one of the most popular benchmarking suites</a> for ML/DL systems. It is comprised of a rich set of system tests that span ML models, software, hardware, and the power efficiency characteristics of an ML/DL system. MLPerf was established in 2018 by several organizations including Google, Harvard University, and Stanford University. Since then, it has greatly boosted innovation in the ML space and has helped vendors deliver high-performance and energy efficient ML/DL systems.The suite includes several benchmarks for different ML tasks. Each benchmark comprises the following characteristics:<ul><li>The name of the task (e.g., Image Classification), which indicates the benchmarking context. For instance, the Image Classification task aims at selecting the class that best describes the content of an image.</li><li>The dataset, which is typically a popular dataset for the task at hand. For instance, the <a href="https://www.image-net.org/" target="_blank">ImageNet database</a> is commonly used for benchmarking Image Classification tasks.</li><li>The model to be used, which identifies the ML/DL model that will be used to gauge the computational performance of the ML hardware and software system. As an example, several variations of the Residual Network (ResNet) are commonly used to benchmark hardware performance in computer vision tasks like Image Classification.</li><li>The Quality Threshold, which denotes a level of acceptable quality for the systems that are benchmarked. For instance, a 74.9% classification accuracy can be used as the threshold for placing an ML/DL system in the Top-1 accuracy.</li></ul>Along with the above-listed parameters, MLPerf standardizes and specifies other aspects of the benchmarking process such as the initialization of parameters, the optimizer schedule to be used, and data augmentation processes. In this way, it ensures that the benchmarked systems are comparable and can be ranked in an unbiased way.&nbsp;The results of the MLPerf benchmarking process are published in a transparent way. This makes it easy for market stakeholders to understand how different systems perform in the various tasks and how they compare to each other. The publication of the results for each system includes information about the system&rsquo;s benchmark scores, its market availability, the models used (e.g., <a href="https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet_v1.md" target="_blank">MobileNet-v1</a> for mobile phones and <a href="https://en.wikipedia.org/wiki/Residual_neural_network" target="_blank">ResNet-50</a> v1.5 for bigger accelerators), as well as the scenarios and configurations of the benchmarking process.&nbsp;The scenarios are destined to cover different deployment configurations, including inference at the edge and inference at cloud data centers. Moreover, there is information about the number of accelerators used by the systems under comparison, as the MLPerf benchmark is designed to rank entire systems and not individual components.In practice, the process of running the MLPerf benchmarking is quite straightforward. It starts with cloning the MLPerf repository and setting up the configuration parameters. Accordingly, the system under benchmarking is added to the list of available system and a proper Docker image for the tests is built. The latter enables data preprocessing, as well as execution of validation models and tests. &nbsp;<h2>Syntiant&rsquo;s Performance in the MLPerf Benchmarking</h2><a href="https://www.eetasia.com/syntiant-leads-tinyml-benchmark-results/" target="_blank">Syntiant is strong leader in tinyML keyword spotting</a> according to the respective MLPerf benchmarking process.&nbsp;&nbsp;The company&rsquo;s performance in the benchmark is summarized as follows:<ul><li>Achieved Performance: Syntiant&rsquo;s NDP120 ran the tinyML keyword spotting benchmark in 1.80 ms. This is a remarkably fast inference, making the company the clear winner in this category.&nbsp;</li><li>Energy Usage: The benchmarked system used 49.59 &mu;J of energy at 1.1V/100 MHz. Energy efficiency was further improved by setting supply voltage down to 0.9 V and reducing clock frequency to 30 MHz, resulting in only 35.29 &mu;J of energy being consumed. Of course, reducing the clock frequency increases the latency to 4.30 ms, but this highlights both the energy and latency floor of the device, leaving it to the system designer to optimize power and performnce based on their specific application needs.&nbsp;</li></ul>The NDP120 is based on Syntiant&rsquo;s second generation AI accelerator core, Syntiant Core 2&reg;, alongside a Tensilica HiFi DSP (Digital Signal Processor) for feature extraction and an Arm Cortex-M0 core for system management. Syntiant did not enter any other benchmark scores for the NDP120, as the chip was architected for time-series data inferences, such as speech and audio events.&nbsp;Syntiant Core 2&reg;, is a highly flexible, ultra-low-power deep neural network inference engine with a highly configurable audio front-end interface. Thanks to this front-end interface, NDP120 is very easily programmable. The ease of deployment is further aided by its native support for all major neural network architectures, allowing for the direct execution of neural network layers without intermediary processing steps.&nbsp;NDP120 can run multiple models, such as a keyword spotting model and an audio event detection model, concurrently and supports flexible feature extraction and acoustic far-field processing with the included DSP. Using multiple neural network layers, the system provides accurate high-performance inference and classification, which supports use cases in a variety of application. For instance, it can accurately capture voice commands (e.g., Alexa prompts) in very noise environments.The device is ideal for ultra-low-power, always-on, edge applications, where an inference based on time series audio or sensor data is required. The deployed solutions using the NDP120 span market &nbsp;verticals in consumer and industrial, such as security devices detecting acoustic events, mobile phones, tablets, smart speakers, smart displays, as well as smart home appliances and security devices. As with all Syntiant Core 2 family of solutions, it provides optimized memory access and at-memory compute, ensuring high efficiency and extremely low power consumption. The latter is clearly evident in the benchmarking of energy consumption across its peers.For more information about Syntiant NDP120 visit: <a href="https://www.syntiant.com/ndp120" rel="noopener noreferrer" target="_blank">https://www.syntiant.com/ndp120</a><h2 id="isPasted">About the sponsor: Syntiant</h2>Syntiant Corp. is a leader in delivering end-to-end deep learning solutions for always-on applications by combining purpose-built silicon with an edge-optimized data platform and training pipeline. The company&rsquo;s advanced chip solutions merge deep learning with semiconductor design to produce ultra-low-power, high-performance, deep neural network processors for edge AI applications across a wide range of consumer and industrial use cases, from earbuds to automobiles. Syntiant&rsquo;s Neural Decision Processors&trade; typically offer more than 100x efficiency improvement, while providing a greater than 10x increase in throughput over current low-power MCU-based solutions, and subsequently, enabling larger networks at significantly lower power.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYzMzU5OTg2NjI5LVVudGl0bGVkIGRlc2lnbi02ICgxKS5wbmciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 300px;" class="fr-fic fr-dib">

Proven benchmarks provide a structured method for comparing ML/DL products and services. 


The Art and Science of Benchmarking Neural Network Architectures

This article was discussed in our <a data-saferedirecturl="https://www.google.com/url?q=https://www.wevolver.com/profile/the.next.byte&source=gmail&ust=1663755892758000&usg=AOvVaw20oAeYG8Fqg7tSb8uYcqUe" href="https://www.wevolver.com/profile/the.next.byte" id="isPasted" rel="noopener noreferrer" target="_blank">Next Byte podcast</a>.&nbsp;<iframe height="200px" width="100%" frameborder="no" scrolling="no" seamless="" src="https://player.simplecast.com/45ace7a2-622d-4f0e-ac28-1fb05fe7efe2?dark=false" class="fr-draggable" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true"></iframe>The full article will continue below.<hr>This article was first published on <a href="https://news.mit.edu/2022/ai-system-makes-models-like-dall-e-2-more-creative-0908" id="isPasted" rel="noopener noreferrer" target="_blank">MIT News</a>. &nbsp;<hr>The internet had a collective feel-good moment with the introduction of DALL-E, an artificial intelligence-based image generator inspired by artist Salvador Dali and the lovable robot WALL-E that uses natural language to produce whatever mysterious and beautiful image your heart desires. Seeing typed-out inputs like &ldquo;smiling gopher holding an ice cream cone&rdquo; instantly spring to life clearly resonated with the world.&nbsp;Getting said smiling gopher and attributes to pop up on your screen is not a small task. DALL-E 2 uses something called a diffusion model, where it tries to encode the entire text into one description to generate an image. But once the text has a lot of more details, it&#39;s hard for a single description to capture it all. Moreover, while they&#39;re highly flexible, they sometimes struggle to understand the composition of certain concepts, like confusing the attributes or relations between different objects.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYzMTU4MDg3NDI5LU1JVC1DU0FJTC1jb21wLWRpZmZ1c2lvbi0wMi1wcmVzcy5qcGciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19" style="width: 766px;" class="fr-fic fr-dib">This array of generated images, showing &ldquo;a river leading into mountains&quot; and &quot;red trees on the side,&rdquo; was generated using a new method developed by MIT researchers. Image courtesy of the researchersTo generate more complex images with better understanding, scientists from MIT&#39;s Computer Science and Artificial Intelligence Laboratory (CSAIL) structured the typical model from a different angle: they added a series of models together, where they all cooperate to generate desired images capturing multiple different aspects as requested by the input text or labels. To create an image with two components, say, described by two sentences of description, each model would tackle a particular component of the image. &nbsp;The seemingly magical models behind image generation work by suggesting a series of iterative refinement steps to get to the desired image. It starts with a &ldquo;bad&rdquo; picture and then gradually refines it until it becomes the selected image. By composing multiple models together, they jointly refine the appearance at each step, so the result is an image that exhibits all the attributes of each model. By having multiple models cooperate, you can get much more creative combinations in the generated images.&nbsp;Take, for example, a red truck and a green house. The model will confuse the concepts of red truck and green house when these sentences get very complicated. A typical generator like DALL-E 2 might make a green truck and a red house, so it&#39;ll swap these colors around. The team&#39;s approach can handle this type of binding of attributes with objects, and especially when there are multiple sets of things, it can handle each object more accurately.&ldquo;The model can effectively model object positions and relational descriptions, which is challenging for existing image-generation models. For example, put an object and a cube in a certain position and a sphere in another. DALL-E 2 is good at generating natural images but has difficulty understanding object relations sometimes,&rdquo; says MIT CSAIL PhD student and co-lead author Shuang Li, &ldquo;Beyond art and creativity, perhaps we could use our model for teaching. If you want to tell a child to put a cube on top of a sphere, and if we say this in language, it might be hard for them to understand. But our model can generate the image and show them.&rdquo;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYzMTU4MTE4NzI4LU1JVC1DU0FJTC1jb21wLWRpZmZ1c2lvbi0wMS1wcmVzc18wLmpwZyIsImVkaXRzIjp7InJlc2l6ZSI6eyJ3aWR0aCI6OTUwLCJmaXQiOiJjb3ZlciJ9fX0=" style="width: 766px;" class="fr-fic fr-dib">This photo illustration was created using generated images from an MIT system called Composable Diffusion, and arranged in Photoshop. Phrases like &ldquo;diffusion model&rdquo; and &ldquo;network&rdquo; were used to generate the pink dots and geometric, angular images. The phrase &ldquo;a horse AND a yellow flower field&rdquo; is included at the top of the image. Generated images of a horse and yellow field appear on the left, and the combined imagery of a horse in a yellow flower field appear on the right. Image: Jose-Luis Olivares, MIT and the researchers<h2>Making Dali proud</h2><a href="https://energy-based-model.github.io/Compositional-Visual-Generation-with-Composable-Diffusion-Models/" target="_blank">Composable Diffusion</a> &mdash; the team&#39;s model &mdash; uses diffusion models alongside compositional operators to combine text descriptions without further training. The team&#39;s approach more accurately captures text details than the original diffusion model, which directly encodes the words as a single long sentence. For example, given &ldquo;a pink sky&rdquo; AND &ldquo;a blue mountain in the horizon&rdquo; AND &ldquo;cherry blossoms in front of the mountain,&rdquo; the team&#39;s model was able to produce that image exactly, whereas the original diffusion model made the sky blue and everything in front of the mountains pink.&nbsp;&ldquo;The fact that our model is composable means that you can learn different portions of the model, one at a time. You can first learn an object on top of another, then learn an object to the right of another, and then learn something left of another,&rdquo; says co-lead author and MIT CSAIL PhD student Yilun Du. &ldquo;Since we can compose these together, you can imagine that our system enables us to incrementally learn language, relations, or knowledge, which we think is a pretty interesting direction for future work.&rdquo;While it showed prowess in generating complex, photorealistic images, it still faced challenges since the model was trained on a much smaller dataset than those like DALL-E 2, so there were some objects it simply couldn&#39;t capture.&nbsp;Now that Composable Diffusion can work on top of generative models, such as DALL-E 2, the scientists want to explore continual learning as a potential next step. Given that more is usually added to object relations, they want to see if diffusion models can start to &ldquo;learn&rdquo; without forgetting previously learned knowledge &mdash; to a place where the model can produce images with both the previous and new knowledge.&ldquo;This research proposes a new method for composing concepts in text-to-image generation not by concatenating them to form a prompt, but rather by computing scores with respect to each concept and composing them using conjunction and negation operators,&rdquo; says Mark Chen, co-creator of DALL-E 2 and research scientist at OpenAI. &ldquo;This is a nice idea that leverages the energy-based interpretation of diffusion models so that <a href="https://energy-based-model.github.io/compositional-generation-inference/" target="_blank">old ideas</a> around compositionality using energy-based models can be applied. The approach is also able to make use of classifier-free guidance, and it is surprising to see that it outperforms the GLIDE baseline on various compositional benchmarks and can qualitatively produce very different types of image generations.&rdquo;&ldquo;Humans can compose scenes including different elements in a myriad of ways, but this task is challenging for computers,&rdquo; says Bryan Russel, research scientist at Adobe Systems. &ldquo;This work proposes an elegant formulation that explicitly composes a set of diffusion models to generate an image given a complex natural language prompt.&rdquo;Alongside Li and Du, the paper&#39;s co-lead authors are Nan Liu, a master&#39;s student in computer science at the University of Illinois at Urbana-Champaign, and MIT professors Antonio Torralba and Joshua B. Tenenbaum. They will present the work at the 2022 <a href="https://twitter.com/eccvconf" target="_blank">European Conference on Computer Vision</a>.<hr>&quot;Reprinted with permission of <a href="https://news.mit.edu/2022/ai-system-makes-models-like-dall-e-2-more-creative-0908" id="isPasted" rel="noopener noreferrer" target="_blank">MIT News</a>&quot;&nbsp;

This array of generated images, showing "a train on a bridge" and "a river under the bridge,” was generated using a new method developed by MIT researchers. Image courtesy of the researchers

Researchers develop a new method that uses multiple models to create more complex images with better understanding.

AI system makes models like DALL-E 2 more creative

Over the past several years, managing one&rsquo;s mental health has become more of a priority with an increased emphasis on self-care. Depression alone affects more than 300 million people worldwide annually. Recognizing this, there is significant interest to leverage popular wearable devices to monitor an individual&rsquo;s mental health by measuring markers such as activity levels, sleep and heart rate.A team of researchers at Washington University in St. Louis and at the University of Illinois Chicago used data from wearable devices to predict outcomes of treatment for depression on individuals who took part in a randomized clinical trial. They developed a novel machine learning model that analyzes data from two sets of patients &mdash; those randomly selected to receive treatment and those who did not receive treatment &mdash; instead of developing a separate model for each group. This unified multi-task model is a step toward personalized medicine, in which physicians design a treatment plan specific to each patient&rsquo;s needs and predict outcome based on an individual&rsquo;s data.&nbsp;Chenyang Lu, the Fullgraf Professor in the McKelvey School of Engineering, led a team including Ruixuan Dai, who worked in Lu&rsquo;s lab as a doctoral student and is now a software engineer at Google; Thomas Kannampallil, associate professor of anesthesiology and associate chief research information officer at the School of Medicine and associate professor of computer science &amp; engineering in McKelvey Engineering; and Jun Ma, MD, PhD, the Beth and George Vitoux Professor of Medicine at the University of Illinois Chicago (UIC), and colleagues to develop the model using data from a randomized clinical trial conducted by UIC with about 100 adults with depression and obesity.&ldquo;Integrated behavioral therapy can be expensive and time consuming,&rdquo; Lu said. &ldquo;If we can make personalized predictions for individuals on whether it is likely a patient would be responsive to a particular treatment, then patients may continue with treatment only if the model predicts their conditions are likely to improve with treatment but less likely without treatment. Such personalized predictions of treatment response will facilitate more targeted and cost-effective therapy.&rdquo;&nbsp;Results of the research were published in the&nbsp;Proceedings of the ACM on Interactive, Model, Wearable and Ubiquitous Technologies and will be presented at the UbiComp 2022 conference in September.&nbsp;In the trial, patients were given Fitbit wristbands and psychological testing. About two-thirds of the patients received behavioral therapy, and the remaining patients did not. Patients in both groups were statistically similar at baseline, which gave the researchers a level playing field on which to discern whether treatment would lead to improved outcomes based on individual data.&nbsp;Clinical trials of behavioral therapies often involved relatively small cohorts due to the cost and duration of such interventions. The small number of patients created a challenge for a machine learning model, which typically performs better with more data. However, by combining the data of the two groups, the model could learn from a larger dataset, which captured the differences in those who had undergone treatment and those who had not. They found that their multi-task model predicted depression outcomes better than a model looking at each of the two groups separately.&ldquo;We pioneered a multi-task framework, which combines the intervention group and the control group in a randomized control trial to jointly train a unified model to predict the personalized outcomes of an individual with and without treatment,&rdquo; said Dai, who earned a doctorate in computer science in 2022. &ldquo;The model integrated the clinical characteristics and wearable data in a multi-layer architecture. This approach avoids splitting the study cohorts into smaller groups for machine learning models and enables a dynamical knowledge transfer between the groups to optimize prediction performance for both with and without intervention.&rdquo;&ldquo;The implications of this data-driven approach extend beyond randomized clinical trials to implementation in clinical care delivery where the ability to make personalized prediction of patient outcomes depending on the treatment received, and to do so early and along the treatment course, could meaningfully inform shared-decision making by the patient and the treating physician in order to tailor the treatment plan for that patient,&rdquo; Ma said.The machine learning approach provides a promising tool to build personalized predictive models based on data collected from randomized controlled trials. Going forward, the team plans to leverage the machine learning approach in a new randomized controlled trial of telehealth behavioral interventions using Fitbit wristbands and weight scales among patients in a weight loss intervention study.<hr>Dai R, Kannampallil R, Zhang J, Lv N, Ma J, Lu C. Multi-Task Learning for Randomized Controlled Trials: A Case Study on Predicting Depression with Wearable Data.&nbsp;Proceedings of the ACM on Interactive, Model, Wearable and Ubiquitous Technologies,6, 2, Article 50 (July 2022).&nbsp;<a href="https://nam10.safelinks.protection.outlook.com/?url=https%3A%2F%2Fdl.acm.org%2Fdoi%2F10.1145%2F3534591&data=05%7C01%7Cmillerbe%40wustl.edu%7Cc0c6b767a9da4bb5219708da8615ec47%7C4ccca3b571cd4e6d974b4d9beb96c6d6%7C0%7C0%7C637969726119601617%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=5ulqE6FvpcsqO5JD2PeNWntx4o9kwGzhq%2B3beQq0kCc%3D&reserved=0" target="_blank">DOI: 10.1145/3534591</a>

Interdisciplinary team builds multi-task machine learning model for randomized controlled trial

Personalized prediction of depression treatment outcomes with wearables

Artificial intelligence (AI) is showing significant promise in medical imaging. To translate this promise to reality requires rigorous evaluation of these algorithms.To develop guidelines to evaluate artificial intelligence (AI) in nuclear-medicine imaging, Richard L. Wahl, MD, director of Mallinckrodt Institute of Radiology (MIR) at Washington University School of Medicine in St. Louis, then president of the Society of Nuclear Medicine and Molecular Imaging (SNMMI), established the SNMMI Artificial Intelligence Task Force, within which an evaluation team was formed. This team, comprised of computational imaging scientists, physicians, physicists, biostatisticians and representatives from industry and regulatory agencies, was led by Abhinav Jha, assistant professor of biomedical engineering in the McKelvey School of Engineering and assistant professor of radiology at MIR.The guidelines proposed by this team, known as the Recommendations for Evaluation of AI for Nuclear Medicine (RELAINCE) guidelines, will be published in the September 2022 issue of the&nbsp;Journal of Nuclear Medicine.&nbsp;&ldquo;Artificial intelligence is showing tremendous promise in medical imaging, specifically in nuclear medicine imaging, in a multitude of applications ranging from image generation, enhancement and analysis,&rdquo; Jha said. &ldquo;We have seen a lot of research in this area, including multiple papers by our own group. However, for clinical translation of these algorithms, rigorous evaluation is needed.&ldquo;Lack of rigorous evaluation may have multiple adverse consequences, including reducing credibility of research findings, misdirection of future research, and, most importantly, yielding tools that are useless or even harmful to patients,&rdquo; Jha continued. &ldquo;In our discussions, it emerged that there is an important need for guidelines to perform such evaluation.&rdquo;As an example, there is significant research in developing AI algorithms for processing nuclear-medicine images acquired at low doses.&ldquo;Along this direction, in our own lab, we developed an algorithm to process low-dose cardiac SPECT images and were excited by the results because the resulting images visually looked great,&rdquo; Jha said. &ldquo;But it&rsquo;s not whether they look great that matters &ndash; it is how they will do on the task that is required from the images, in this case, detecting cardiac defects. And on that task, the algorithm performed no better compared to the original low-dose images.&ldquo;We found that while the images looked good, in some cases, they removed the lesions, and in others, they introduced false lesions,&rdquo; he said. &nbsp;None of those results is acceptable, showing the need to evaluate the algorithms based on the clinical task.The task force proposed that all AI algorithms should be evaluated on clinical tasks, and the evaluations should yield a claim that defines the clinical task for which the algorithm was evaluated, demographics, imaging procedures, and process to extract task-specific information that was used in the evaluation study, and quantitative figures of merit to evaluate performance on clinical task.&ldquo;A proper and clear definition of a claim indicating the intended use and the validity of the AI algorithm is of utmost importance, and the claim should be substantiated with a proper and extensive evaluation of the AI method,&rdquo; said Ronald Boellaard, professor of radiology and nuclear medicine at Amsterdam University Medical Centers and senior author of the paper. &ldquo;The claim should specifically indicate under which conditions and for which patients the algorithm can be used as well as any limitations or factors that could result in incorrect or less accurate performances.&rdquo;Wahl, the Elizabeth E. Mallinckrodt Professor of Radiology and a co-author, said the AI task force has generated several important papers which will help move AI methods forward from research to clinical practice in nuclear medicine.&ldquo;AI has the potential to spread expertise globally, but if not implemented properly, could spread inaccuracies,&rdquo; he said. &ldquo;Thus, the work by Dr. Jha and colleagues on the RELAINCE criteria is so important to assuring valid population and task-specific AI methods are developed and deployed.&rdquo;The task force proposed a framework with four classes to evaluate the algorithms for promise, technical task-specific efficacy, clinical decision making and post-deployment efficacy. The intent behind this framework is that it will guide AI developers conduct the evaluation study that provides evidence to support their intended claim. Additionally, the task force proposed best practices to evaluate the AI algorithms for each of the four classes.&ldquo;We want to make sure these algorithms are evaluated well so that they can assist well in clinical tasks, and hence the patients get the best treatment,&rdquo; Jha said. &ldquo;We want AI to help, not hurt the patient.&rdquo; &nbsp;&nbsp;&ldquo;It was an honor and a privilege to be part of this evaluation team, working with world-renowned researchers getting together to address this important need. A memory I will cherish!&rdquo; Jha added.<hr>Jha AK, Bradshaw TJ, Buvat I, Hatt M, KC P, Liu C, Obuchowski NF, Saboury B, Slomka PJ, Sunderland JJ, Wahl RL, Yu Z, Zuehlsdorff S, Rahmim A, Boellaard R. Nuclear Medicine and Artificial Intelligence: Best Practices for Evaluation (the RELAINCE guidelines).&nbsp;Journal of Nuclear Medicine, Aug. 1, 2022. DOI: 10.2967/jnumed.121.263239

To develop guidelines to evaluate artificial intelligence (AI) in nuclear-medicine imaging, an interdisciplinary team was set up within an AI taskforce established by Richard L. Wahl, MD, director of Mallinckrodt Institute of Radiology (MIR) at Washington University School of Medicine in St. Louis. The team, led by Abhinav Jha, assistant professor of biomedical engineering in the McKelvey School of Engineering and assistant professor of radiology at MIR, has published Recommendations for Evaluation of AI for Nuclear Medicine (RELAINCE) guidelines.

Artificial intelligence (AI) is showing significant promise in medical imaging. To translate this promise to reality requires rigorous evaluation of these algorithms.

Evaluation of AI for medical imaging: A key requirement for clinical translation

Hospitals spend about one-third of their expenses on perioperative care &ndash; the high-stakes period just before and after a patient is in surgery &mdash; to prevent complications afterward. Washington University in St. Louis researchers have developed a machine learning approach that exploits the large amount of clinical data collected during perioperative care to predict potential post-operative complications, which could ultimately improve patient outcome and reduce the cost of care.Chenyang Lu, the Fullgraf Professor in the McKelvey School of Engineering, and Bing Xue, a doctoral student in Lu&rsquo;s lab, collaborated with a team of physicians and researchers from Washington University School of Medicine to develop machine learning models that provided superior performance for predicting post-operative complications and surgery duration. They presented their results at the ACM SIGKDD Conference on Knowledge Discovery and Data Mining.&nbsp;When a patient is prepared for and undergoes surgery, health care providers gather a wealth of information comprising hundreds of variables such as demographics, history of comorbidities, lab tests, medications, clinical notes and physiological signals. While a large number of variables, or high dimensionality of data, provides rich information that can be learned with a machine learning model, the high dimensionality of perioperative data also carries several challenges for the model: The high dimensionality of data may lead to a &ldquo;brittle&rdquo; model, which fits the training data well, but suffers deterioration and instability in predictive performance in the real world. In addition, perioperative data have highly variable missing rates across different variables depending on the characteristics of the patient and the specific surgery. Given the critical nature of perioperative model, it is imperative for the model to maintain &ldquo;robust&rdquo; predictive performance despite the complexity of perioperative data.&nbsp;To deal with the complexity of perioperative data, Lu and Xue developed a deep learning approach called clinical variational autoencoder (cVAE) to transform the data to latent representations with lower dimensionality while capturing the hidden nonlinear relationship of the original perioperative variables. Using data sets from hospitals within the BJC HealthCare system in St. Louis they trained two models &mdash; one for predicting patients who would develop postoperative delirium, and one for predicting the duration of surgeries. Using their cVAE model, Xue transformed 562 clinical variables into 10 variables that captured essentially the same information. This representation of compressed data, or latent space, can then be used to make robust predictions of surgical outcomes, Lu and Xue said.&ldquo;At Washington University, we are integrating state-of-the-art machine learning approaches, sophisticated real-time telemedicine support and human-in-the loop adjudication to advance efficiency, quality and safety for thousands of patients receiving complex perioperative care,&rdquo; said Michael Avidan, MBBCh, the Dr. Seymour and Rose T. Brown Professor and head of the Department of Anesthesiology at Washington University School of Medicine.&ldquo;This model shows us how to change data into information then into knowledge and affect clinical decisions,&rdquo; said Joanna Abraham, associate professor of anesthesiology in the School of Medicine and a co-author on the paper. &ldquo;A human cannot process all of this information and find the patterns. Adding the additional insight is useful for making better decisions.&rdquo;While other variational autoencoders exist, they have several issues.&ldquo;Traditional variational autoencoders have a performance limitation,&rdquo; Lu said. &ldquo;After you encode all these variables into a compact representation, the predicted performance tends to suffer.&rdquo;&ldquo;When they do such auto-encoding, other existing models will ignore the labels or the prediction tasks information,&rdquo; Xue said. &ldquo;In essence, the transformed latent space is not going to maximize the prediction performance because we&#39;re not making use of that part of information.&rdquo;&ldquo;In contrast, cVAE dedicated a dimension to prediction in the latent space: What is the probability of this patient to develop delirium after surgery? How long will this surgery last?&rdquo; Lu said. &ldquo;When we transform the data set, we use the outcome to guide transformation, and the resulting representation captures the prediction. There is no drop in predictive performance when you build a model on the representation you learned.&rdquo;&nbsp;Also, there is an issue with interpretability of the data representation. Clinicians need to interpret the nonlinear encoding and trust the predictions. To preserve interpretability, cVAE groups patients into interpretable phenotypes in the latent space by disentangling the latent representation, Lu explained.&ldquo;The latent representation created with cVAE therefore enables us to build robust, accurate and interpretable predictive models for perioperative care and other challenging applications with high-dimensional and noisy data,&rdquo; he said.The model is an entry to bringing machine learning into clinical care, the researchers said.&ldquo;Decision support derived from machine learning and AI provides enormous potential benefit to perioperative clinicians, yet the opaque nature of predictive algorithms oftentimes is a barrier to implementation of this technology,&rdquo; said York Jiao, MD, assistant professor of anesthesiology. &ldquo;In this article, interpretable organization of the latent space is demonstrated without meaningful loss of model performance. This achievement may increase trust amongst clinicians who utilize predictive algorithms in their workflow and helps pave the way for operationalization of AI in health care.&rdquo;&nbsp;<hr>Xue B, Jiao Y, Kannampallil T, Fritz B, King C, Abraham J, Avidan M, Lu C. Perioperative Predictions with Interpretable Latent Representation. Association for Computing Machinery SIGKDD Conference on Knowledge Discovery and Data Mining, Aug. 14-18. DOI:<a href="https://dl.acm.org/doi/10.1145/3534678.3539190" id="isPasted" target="_blank">10.1145/3534678.3539190</a>

Chenyang Lu, the Fullgraf Professor in the McKelvey School of Engineering, and Bing Xue, a doctoral student in Lu’s lab, collaborated with a team of physicians and researchers from Washington University School of Medicine to develop machine learning models that provided superior performance for predicting post-operative complications and surgery duration. (iStock photo)

Interdisciplinary team tackles complexity of clinical data in perioperative care

Predicting surgical outcomes with machine learning

A robot, or a group of robots, on a mission often have conflicting priorities. &nbsp;The primary mission of a team of search and rescue robots, for example, is to find survivors. But when they do, they also need to be able to send information to the rest of the team. The best way to find survivors may be to spread out, but the best way to send information is to stay close together. If robotic teams are going to be deployed in more real-world situations, they&rsquo;re going to need a system to balance these priorities.Computer scientists at the&nbsp;<a href="https://www.seas.harvard.edu/" target="_blank">Harvard John A. Paulson School of Engineering and Applied Sciences</a> (SEAS) have developed a new control mechanism for robotic teams that allow them to automatically calculate the tradeoffs of maintaining a resilient communication network and achieving their primary goal of moving through an environment. The team calculated which types of environments a robotic team could navigate while maintaining communication and which environments would force a team to break communication.&ldquo;We wanted to understand and analyze the tradeoff between making progress and maintaining a resilient communication network and build a control mechanism that can deal with those tradeoffs without deadlocking the robotic team,&rdquo; said&nbsp;<a href="https://gil.seas.harvard.edu/" target="_blank">Stephanie Gil, Assistant Professor of Computer Science at SEAS and senior author of the paper.</a><a href="https://gil.seas.harvard.edu/" target="_blank">The research was presented recently at the&nbsp;</a><a data-lf-fd-inspected-kn9eq4roawkarlvp="true" href="http://www.roboticsproceedings.org/rss18/p053.pdf" target="_blank">Robotics: Science and Systems</a> conference, where it was selected as a finalist for best paper.In multi-robot systems, communication is often dependent on the distance between individual robots. Many robotic teams use networked communication, relaying signals from one robot to another to cover increased distance between team members.&ldquo;A resilient communication network has multiple paths of communication between any robot on the team to any other and so even if multiple robots were to fail, there are other ways to pass the information,&rdquo; said Matthew Cavorsi, a graduate student at SEAS and first author of the paper.<iframe src="https://www.youtube.com/embed/HPdAcipt9K8?&wmode=opaque&rel=0&enablejsapi=1&origin=https://www.wevolver.com" frameborder="0" allowfullscreen="" class="fr-draggable" data-gtm-yt-inspected-12="true" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true" style="width: 767px; height: 427px;"></iframe>SEAS researchers developed a new control mechanism for robotic teams maintaining a resilient communication network while moving through an environment.&nbsp;These resilient robotic systems need to maintain specific formations with certain distances between each robot to keep those multiple paths of communications open. But what happens when an environment, such as a narrow hallway, constricts the formation? If the robotic team is forced into a single-file line, the resiliency of the network breaks down because there is only one path along which to send a signal.&ldquo;If the communication network presents single points of failure, then this multi-robot system becomes very fragile,&rdquo; said Lorenzo Sabattini, Associate Professor of Robotics at the University of Modena and Reggio Emilia in Italy and co-author of the paper. &nbsp;&ldquo;Our work aims at removing this issue by determining when the system can maintain resiliency and achieve its goal, when resiliency and the goal are incompatible, and how the system can regain resiliency as quickly as possible after having to temporarily break it, thus paving the way towards reliable real-world application of multi-robot systems.&rdquo;The team identified a cut-off point where the robotic team is connected enough to be resilient. Through models and experiments, the team demonstrated that the control mechanism can compute the connectedness of the team as it moves and makes adjustments to keep it within the threshold.Cavorsi developed two types of controls: one where resiliency is a soft constraint and one where resiliency is a hard constraint.If a multi-robot system with resiliency as a soft constraint encounters an environment that would force it to drop below the resiliency threshold, such as a narrow hallway or large boulder, it will break resiliency long enough to move through the environment. If resiliency is a hard constraint, the system will turn back or find another way through the environment that doesn&rsquo;t require it to break resiliency.Soft constraint resiliency could work well for search and rescue robots where achieving the goal of finding survivors is worth the risk of temporarily weakening their communications. Resiliency could be more important for robotic systems on Mars, for example, where maintaining communications and avoiding dangerous obstacles is critical to the mission.<iframe src="https://www.youtube.com/embed/E7E59Mm-pS0?&wmode=opaque&rel=0&enablejsapi=1&origin=https://www.wevolver.com" frameborder="0" allowfullscreen="" class="fr-draggable" data-gtm-yt-inspected-12="true" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true" style="width: 767px; height: 418px;"></iframe>Multi-robot teams may one day provide assistance to traditional rescue efforts.&nbsp;The research was co-authored by Beatrice Capelli and Lorenzo Sabattini.&nbsp;

A new control mechanism allows teams of robots (above) to calculate the tradeoffs of maintaining a communication network and move through difficult terrain, including narrow hallways. (Credit: REACT Lab/Harvard SEAS) 

Resilient robot teams juggle competing priorities without deadlock

Resilient robots

In the time it takes you to read this blog, the average adult human will have made more than 200 decisions, or so it has been claimed. Columbia University research posits a far more&nbsp;modest total of 70 conscious daily decisions. Whatever the actual number, humans make a lot of decisions, and we&rsquo;ve become highly skilled at it.We&#39;re good at decision-making because we&#39;ve learned to compensate for our weaknesses. We use mental shortcuts or heuristics to counteract our inability to process lots of information. We use gut feeling and intuition to help us make decisions based on unconscious recall of observation and experience.<h2>Thinking like a machine</h2>Machines, on the other hand, &#39;think&#39; very differently, relying on their computational superiority to rapidly process hundreds of millions of data points and make decisions based on consequence rather than intuition. As such, computers can now do some specific tasks very well. They have, for example, comfortably conquered the best human players in games such as chess and Go. More usefully, they can also identify anomalies in thousands of x-rays faster and more reliably than any radiologist, although the difference is marginal.<h2>Working together in the IoT</h2>For now, at least, machines need humans as much as humans need machines. At this intersection, machine learning (ML) offers intriguing possibilities for managing the billions of end devices that comprise the IoT. ML is a practical, mathematical field. It involves humans creating algorithms and programs that allow computers to follow hundreds of millions of instructions. The computers look for occasional deviations in an almost endless stream of otherwise unremarkable data&mdash;and then interpret patterns in those deviations to solve real-world problems.<h2>The power of machine learning</h2>Picture the smart city of the future, for example. Streetlights will carry sensors monitoring traffic flow and air quality. Weather stations will measure meteorological variables to predict hazardous climate events. Buildings will be embedded with hundreds of sensors to monitor everything from air conditioning and lifts to water and gas pipes. These devices and millions more will generate trillions of gigabytes of data. Making sense of it is a major headache that engineers believe ML can help them relieve.Relaying that volume of data to the cloud so that it can be crunched, is untenable. The often battery-powered end devices don&#39;t have the energy to do it, the networks can&#39;t carry that kind of load, and even if they could, the cost to do so would be prohibitive.<h2>TinyML at the edge</h2>However, today&#39;s IoT devices can support TinyML, a scaled-down version of ML that enables edge devices to monitor data and use algorithms to detect deviations constantly. Data gathered from one or multiple sensors is given to the cloud, with powerful cloud servers then able to train an ML model that is updated over the air to the edge device. This enables the end device to determine what data to send to the cloud in the future, if a data deviation suggests an air conditioning unit is going to break down or a tornado is on its way, for example. In the future, the &lsquo;learning&rsquo; will be performed on the device, but we are not there yet.<h2>ML augmenting human intelligence</h2>Research on the possibilities of augmented intelligence is still in its infancy, as is how it might be applied to the IoT where large-scale problems cannot always be solved by either computers or humans alone. This collaborative intelligence could allow both parties to play to their strengths. Humans must train machines and their ML algorithms to perform the work they are designed to do. Humans are also essential to interpret a machine&#39;s decisions and ensure they are functioning properly, safely, and responsibly.A machine, on the other hand, can boost human analytical and decision-making abilities by providing the right information at the right time, and filtering out any noise it deems unnecessary for us to make our decision. Machines and their ML algorithms can also be used as an instructional tool to improve the skills and performance of humans.<h2>Who is in control with ML?</h2>But the machines have not yet risen, and for now it is us, not them, calling the shots. The decision to entrust autonomous decision-making to IoT solutions without human oversight must be carefully weighed based on performance, cost, and risk factors.<hr>This article was first published on Nordic&#39;s&nbsp;<a href="https://blog.nordicsemi.com/getconnected" target="_blank" rel="nofollow" id="isPasted"></a><a href="https://blog.nordicsemi.com/getconnected?utm_campaign=2021%20wevolver" rel="nofollow" target="_blank">Get Connected Blog</a>. &nbsp;

For now, at least, machines need humans as much as humans need machines. At this intersection, machine learning (ML) offers intriguing possibilities for managing the billions of end devices that comprise the IoT. ML is a practical, mathematical field.

Harnessing man and machine's strengths for better IoT outcomes with ML

This article is a part of our <a href="http://www.wevolver.com/student-program" target="_blank">University Technology Exposure Program</a>. The program aims to recognize and reward innovation from engineering students and researchers across the globe.#1. <a href="https://www.wevolver.com/article/conceptual-design-of-smart-multi-farm-produce-dehydrator-using-a-low-cost-plc-and-raspberry-pi" target="_blank">Conceptual design of smart multi-farm produce dehydrator using a low-cost PLC and Raspberry Pi</a>&nbsp; By&nbsp;<a href="https://www.wevolver.com/profile/sunkanmi.oluwaleye" target="_blank">Sunkanmi Oluwaleye</a>Processing food/crop often requires removing moisture content without affecting their nutritional values. To carry out the process effectively, a special dehydrator/oven is required that can evenly distribute the generated heat and remove all moisture.This student article by Sunkanmi proposes a low-cost raspberry pi and programmable logic controller (PLC) based dehydrator that gives the users the flexibility to control both the drying chamber&rsquo;s temperature and humidity from its web interface via a mobile device or the dehydrator&rsquo;s human machine interface (HMI).Read the complete write-up <a href="https://www.wevolver.com/article/conceptual-design-of-smart-multi-farm-produce-dehydrator-using-a-low-cost-plc-and-raspberry-pi" target="_blank">here</a>.<hr>#2. <a href="https://www.wevolver.com/article/new-image-recognition-technique-for-intuitive-understanding-in-class-of-the-dynamic-response-of-high-rise-buildings" target="_blank">New Image Recognition Technique for Intuitive Understanding in Class of the Dynamic Response of High-Rise Buildings</a> By&nbsp;<a href="https://www.wevolver.com/profile/rocio.porras.soriano" target="_blank">Rocio Porras Soriano</a>Structural engineering is a subject of civil engineering that requires students to develop an intuition of the structural response of different construction designs. In this article, Rocio explains how complementing traditional master classes with additional activities can improve students&rsquo; learning experiences.An image-recognition algorithm is proposed to identify the dynamic response (deflections, velocity, and accelerations) of small-scale buildings on an educational shaking table. Students are challenged to improve the seismic behavior of a reference small-scale building. Then, the proposed algorithm is used to compare the different typologies proposed. A survey conducted at the end of the course indicates an improvement in the student&#39;s confidence and motivation.Read more about the methodology and the results <a href="https://www.wevolver.com/article/new-image-recognition-technique-for-intuitive-understanding-in-class-of-the-dynamic-response-of-high-rise-buildings" target="_blank">here</a>.<hr>#3. <a href="https://www.wevolver.com/article/a-machine-learning-approach-for-an-improved-inertial-navigation-system-solution" target="_blank">A Machine Learning Approach for an Improved Inertial Navigation System Solution</a> By&nbsp;<a href="https://www.wevolver.com/profile/ashraf.abosekeen" target="_blank">Ashraf Abosekeen</a>An inertial Navigation System (INS) is a component of mechanical devices that uses the data collected from sensors like accelerometers and gyroscopes to plan movements. The system depends on the accuracy of the Inertial Measurement Unit (IMU), which provides INS with the required data. IMUs used in low-cost micro-electro-mechanical systems (MEMSs) suffer from huge error sources such as bias, the scale factor, scale factor instability, and highly non-linear noise.This article by Ashraf proposes a machine-learning-based adaptive neuro-fuzzy inference system (ML-based-ANFIS) that improves the traditional INS by up to 70% in 2D positioning and 92% in 2D velocity.Visit <a href="https://www.wevolver.com/article/a-machine-learning-approach-for-an-improved-inertial-navigation-system-solution" target="_blank">this link</a> for all details.<hr>#4. <a href="https://www.wevolver.com/article/combining-sla-3d-printing-and-soft-lithography-for-fast-versatile-and-accessible-high-resolution-fabrication-of-customised-multiscale-cell-culture-devices-with-complex-designs" target="_blank">Combining SLA 3D printing and soft lithography for fast, versatile, and accessible high-resolution fabrication of customised multiscale cell culture devices with complex designs</a> By&nbsp;<a href="https://www.wevolver.com/profile/cathleen.hagemann" target="_blank">Cathleen Hagemann</a>Taking advantage of low-cost, high-resolution desktop resin 3D printers combined with Polydimethylsiloxane (PDMS) soft-lithography, this article explains the optimized microfabrication pipeline capable of generating a wide variety of customizable devices for cell culture and tissue engineering in an easy, fast reproducible way for a fraction of the cost of conventional microfabrication or commercial alternatives.&nbsp;This technique enables the manufacture of complex devices across scales bridging the gap between microfabrication and fused deposition molding (FDM) printing.<a href="https://www.wevolver.com/article/combining-sla-3d-printing-and-soft-lithography-for-fast-versatile-and-accessible-high-resolution-fabrication-of-customised-multiscale-cell-culture-devices-with-complex-designs" target="_blank">Here</a> is the link to the student article.<hr>#5. <a href="https://www.wevolver.com/article/spatio-temporal-analysis-of-archived-web-news-for-precise-political-event-detection-and-impact-analysis-in-indias-southern-states" target="_blank">Spatio-Temporal Analysis of Archived Web News for Precise Political Event Detection and Impact Analysis in India&#39;s Southern States</a> By&nbsp;<a href="https://www.wevolver.com/profile/pratik.dhabal.deo" target="_blank">Pratik Dhabal Deo</a>A nation&rsquo;s political system directly impacts its economy. Every political party has its own economic policies, which have a major impact on the economic stability of the nation.This student article by Pratik explains Precise Political Event Detection using Spatio-Temporal Analysis (PPED-STA) schemes proposed by political event mapping through the extraction of data from the archives of web news. The aim is to use machine learning algorithms to predict precisely the possibilities of political events.Read the submission <a href="https://www.wevolver.com/article/spatio-temporal-analysis-of-archived-web-news-for-precise-political-event-detection-and-impact-analysis-in-indias-southern-states" target="_blank">here</a>.<hr>#6. <a href="https://www.wevolver.com/article/artificial-intelligence-assisted-architected-materials-in-additive-manufacturing" target="_blank">Artificial Intelligence Assisted Architected Materials in Additive Manufacturing</a> By&nbsp;<a href="https://www.wevolver.com/profile/mohammad.mirzaali" target="_blank">Mohammad Mirzaali</a>The current advancement in multi-material additive manufacturing technologies has provided unique opportunities to create 3D printed structures with not only complex geometries at several length scales but arbitrary deposition of multiple materials (i.e., soft or hard) in 3D space.In this article, Mohammad Mirzaali shares the design approaches and (bio)fabrication techniques for creating (bio)materials with tailor-made properties. Two frontier technologies, namely multi-material 3D printing and AI, are used to create various types of &ldquo;digital materials&rdquo; with rare properties.Access the full article <a href="https://www.wevolver.com/article/artificial-intelligence-assisted-architected-materials-in-additive-manufacturing" target="_blank">here</a>.<hr>#7. <a href="https://www.wevolver.com/article/dot-hazmat-an-android-application-for-detection-of-hazardous-materials-hazmat" target="_blank">DOT-HAZMAT: An Android application for detection of hazardous materials (HAZMAT)</a> By&nbsp;<a href="https://www.wevolver.com/profile/mihir.pamnani" target="_blank">Mihir Pamnani</a>HAZMAT signs play a crucial role in rescue operations and hence an accurate detection tool is required in cases of accidents caused by hazardous materials. There are several approaches and methods for HAZMAT sign detection but the ones that use deep learning are the most successful ones.In this submission, Mihir explains DOT-HAZMAT, an Android application based on Machine Learning and Computer Vision that locates and categorizes HAZMAT placards in perilous rescue operations. The article explains the enormous potential of neural networks and open-source libraries with computer vision for solving problems of extreme importance.Check out all the details by visiting <a href="https://www.wevolver.com/article/dot-hazmat-an-android-application-for-detection-of-hazardous-materials-hazmat" target="_blank">this link</a>.<hr>#8. <a href="https://www.wevolver.com/article/optimization-of-mechanical-design-bladeless-wind-turbine-for-electricity-fulfillment-in-nusa-tenggara-timur-indonesia" target="_blank">Optimization of Mechanical Design Bladeless Wind Turbine for Electricity Fulfillment in Nusa Tenggara Timur, Indonesia</a> By&nbsp;<a href="https://www.wevolver.com/profile/muhammad.farhan.ramadhany" target="_blank">Muhammad Farhan Ramadhany</a>In this article, Muhammad Farhan Ramadhany explains the concept of Bladeless Wind Turbine wind power plants that can work optimally in areas with low wind speeds (3 to 8 m/s). The article explores how the design can be optimized to work in places with even lower wind availability.Using the optimized designs, electricity demands in areas with low wind speeds can be met. The author has considered the case of Nusa Tenggara Timur, Indonesia, as an example of a region with low wind speed.Here is the <a href="https://www.wevolver.com/article/optimization-of-mechanical-design-bladeless-wind-turbine-for-electricity-fulfillment-in-nusa-tenggara-timur-indonesia" target="_blank">link</a> to the complete article.<hr>#9. <a href="https://www.wevolver.com/article/high-performance-amorphous-carbon-coated-lithium-nickel-manganese-cobalt-oxide-nmc622-cathode-material-with-improved-capacity-retention-for-lithium-ion-batteries" target="_blank">High-Performance Amorphous Carbon Coated Lithium-Nickel-Manganese-Cobalt-Oxide (NMC622) Cathode Material with Improved Capacity Retention for Lithium-Ion Batteries</a> By&nbsp;<a href="https://www.wevolver.com/profile/anish.raj.kathribail" target="_blank">Anish Raj Kathribail</a>Ni-rich NMC has attracted great attention in the battery community due to a combination of high reversible capacity and high operating voltage that stems from two-dimensional lithium-ion diffusion and good lithium-ion conductivity. Coating conducting polymers onto active cathode materials has been proven to mitigate issues at high current densities stemming from the limited conducting abilities of the metal-oxides.However, in spite of these advantages, the material is known to possess problems such as surface side reactions and chemical instability at the highly de-lithiated stages. This student article shows the synergetic effects of a furfuryl polymer coating and subsequent calcination leading to improved electrochemical performance of nickel-rich NMC622.Check out the article <a href="https://www.wevolver.com/article/high-performance-amorphous-carbon-coated-lithium-nickel-manganese-cobalt-oxide-nmc622-cathode-material-with-improved-capacity-retention-for-lithium-ion-batteries" target="_blank">here</a>.<hr>#10. <a href="https://www.wevolver.com/article/fexgan-meta-facial-expression-generation-with-meta-humans" target="_blank">FEXGAN-META: Facial Expression Generation With Meta Humans</a> By&nbsp;<a href="https://www.wevolver.com/profile/j.rafid.siddiqui" target="_blank">J Rafid Siddiqui</a>The subtleness of human facial expressions and a large degree of variation in the level of intensity to which a human expresses them is what makes it challenging to robustly classify and generate images of facial expressions. The lack of good quality data can hinder the performance of deep learning models utilizing this data.This student article by J Rafid Siddiqui proposes a Facial Expression Generation method for Meta-Humans (FExGAN-Meta) that deals with the challenge of robustly classifying and generating images of facial expressions while maintaining subtleness and intensity.Read the article by visiting <a href="https://www.wevolver.com/article/fexgan-meta-facial-expression-generation-with-meta-humans" target="_blank">this link</a>.<hr>#11. <a href="https://www.wevolver.com/article/dynamic-simulation-and-optimization-of-anaerobic-digestion-processes-using-matlab" target="_blank">Dynamic simulation and optimization of anaerobic digestion processes using MATLAB</a> by&nbsp;<a href="https://www.wevolver.com/profile/prabakaran.ganeshan" target="_blank">Prabakaran Ganeshan</a>This student article explains a dynamic model that can predict biomethane production with time series developed based on a modified hill&rsquo;s model using MATLAB.This model predicts the biomethane production for both batch and continuous processes across substrates and under diverse conditions such as total solids, loading rate, and days of operation. The deviation between the literature and the developed model is less than &plusmn; 7.6%, which shows the accuracy and robustness of this model.Visit <a href="https://www.wevolver.com/article/dynamic-simulation-and-optimization-of-anaerobic-digestion-processes-using-matlab" target="_blank">this link</a> to read the submission.<hr>#12. <a href="https://www.wevolver.com/article/quasioptics-for-increasing-the-beam-efficiency-of-wireless-power-transfer-systems" target="_blank">Quasioptics for Increasing the Beam Efficiency of Wireless Power Transfer Systems</a> By Pereira RicardoThe highest beam efficiency in a wireless power transfer (WPT) system that uses focusing components was 51%. Achieved by William Brown in 1964, it used a ~ 3 m diameter reflector for a transfer distance of 7.62 m. This record was unbeaten for 58 years until now, as this student article presents a system that surpasses Brown&rsquo;s work by 25%.&nbsp;Read the article summarizing the research work <a href="https://www.wevolver.com/article/quasioptics-for-increasing-the-beam-efficiency-of-wireless-power-transfer-systems" target="_blank">here</a>.<hr>Submissions received throughout the program are under review by the jury. Stay tuned to know who makes it to the list of winners.Till then, check out the featured UTEP articles by visiting these links:<ul><li dir="ltr"><a href="https://www.wevolver.com/article/utep-interview-building-the-worlds-fastest-hyperloop-tunneling-robot-with-swissloop-tunneling" target="_blank">UTEP Interview: Building the Worlds Fastest Hyperloop Tunneling Robot with Swissloop Tunneling</a></li></ul>Here are the articles summarizing the previous UTEP submissions:<ul><li dir="ltr"><a href="https://www.wevolver.com/article/university-technology-exposure-program-2022-june-update" target="_blank">University Technology Exposure Program 2022: June Update</a></li><li dir="ltr"><a href="https://www.wevolver.com/article/university-technology-exposure-program-2022-may-update" target="_blank">University Technology Exposure Program 2022: May Update</a></li><li dir="ltr"><a href="https://www.wevolver.com/article/university-technology-exposure-program-2022-april-update" target="_blank">University Technology Exposure Program 2022: April Update</a></li><li dir="ltr"><a href="https://www.wevolver.com/article/university-technology-exposure-program-2022-march-update" target="_blank">University Technology Exposure Program 2022: March Update</a></li><li dir="ltr"><a href="https://www.wevolver.com/article/university-technology-exposure-program-2022-february-update" target="_blank">University Technology Exposure Program 2022: February Update</a></li></ul><h2 dir="ltr">About the University Technology Exposure Program 2022</h2>Wevolver, in partnership with <a href="https://mouser.com/" target="_blank">Mouser Electronics</a> and <a href="https://www.ansys.com/academic" target="_blank">Ansys</a>, is excited to announce the launch of the University Technology Exposure Program 2022. The program aims to recognize and reward innovation from engineering students and researchers across the globe. Learn more about the program <a href="http://www.wevolver.com/student-program" target="_blank">here</a>.<img alt="university-technology-exposure-program-mouser-ansys" src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjYwMTI3NzY2ODc0LTE2NjAxMjc3NjY4NzQucG5nIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" class="fr-fic fr-dii">

UTEP 2022 ends with 12 submissions in the final month. The recent articles covered different areas of technology like smart farming, AI, 3D printing, wind energy, structural engineering, biotechnology, and more. As we eagerly wait for the results, here is an article summarizing the new submissions.

University Technology Exposure Program 2022: July-August Final Update

60% of those surveyed reported that AI was yielding significant cost savings, with the greatest benefits being shown in manufacturing and supply chain functions. Two years on, KPMG reported that 93% of their respondents in industrial manufacturing had already implemented moderate or fully functional AI into their processes. AI adoption (ML in particular), was visible in several main areas:<ul><li>Predictive maintenance</li><li>Defects monitoring</li><li>Network planning</li><li>Detailed scheduling</li><li>Forecasting accuracy</li></ul><h2>Predictive Maintenance</h2>One of the biggest noticeable impacts made by AI and ML is in&nbsp;<a href="https://www.rowse.co.uk/blog/post/what-is-predictive-maintenance" target="_blank" title="What Is Predictive Maintenance">predictive maintenance</a>. Unexpected machine downtime is a normal, and usually expensive, expectation for most manufacturing businesses. A sudden breakdown can lead to increased unit costs, production delays &ndash; and paying premium freight charges so that orders can be rushed out once your factory is functional again.Predictive maintenance does away with all that. Predictive AI algorithms collect live data from sensors on your factory floor, and this big data is analysed to help increase efficiency. It starts by identifying locations or circumstances where your equipment&#39;s useful life might be diminished. ML is also able to correlate specific factors, indecipherable to any human planner, with impending breakdowns. When these factors are identified and programmed, you&#39;ll be alerted about an impending breakdown. You can then take proactive measures to determine the least disruptive moment when maintenance can be carried out. In this instance, ML provides early identification of maintenance requirements before your factory needs a complete shutdown.In AI-enabled production environments, planners have a better picture of what to expect in areas like demand and capacity restrictions. Many industrial manufacturers have already begun implementing AI tech into their maintenance operations, and a significant proportion believes that preventive maintenance will be the strongest factor in their company&rsquo;s future success. Many early adopters succeeded during COVID-19 because of their AI capabilities, while other companies are planning to implement more ML so they can catch up.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1659945063828-1659945063828.png" alt="How To Use AI/ML To Optimise Manufacturing Costs" class="fr-fic fr-dii"><h2>Defects Monitoring</h2>AI is contributing to the enhancement of production quality control. Advanced connectivity includes installing sensors within or above the production line. These can identify and correct process defects in real time, instead of waiting until the production process is complete. In this way, you can significantly reduce production line downtime and wastage. ML can leverage the data to analyse the results of any defects for trends or patterns which could help to identify the root cause. As much of the overall manufacturing cost of your products is in source materials, this capability could make a significant difference to your operations. Minimising scrap materials and rework will also help to cut down costs and increase profit margins.<h2>Network Planning</h2>The principal aim of AI/ML adoption is optimisation. These technological innovations are all designed to streamline and speed up processes,&nbsp;<a href="https://www.rowse.co.uk/blog/post/10-ways-to-reduce-manufacturing-costs" target="_blank" title="10 Ways To Reduce Manufacturing Costs">reduce production costs</a> and deliver a better quality end result. Reducing the immediate costs of materials and labour is the first step to optimisation.The next is to optimise the entire production network. This involves deploying algorithms, as before, to identify trends and patterns that couldn&rsquo;t possibly be discerned by human eyes. This time, though, it&#39;s on a larger scale, looking at the big picture. What you want is to find more efficient, less wasteful ways to structure not only your factory floor, but your whole production network. The goal of a more efficient network is a smoother overall operation and the effective reduction of logistics costs. In addition, you&#39;ll cut down on any replanning work that might be caused by a mechanical process or supply chain disruption. Your operation will become progressively leaner, with the knock-on effect of reducing your future capital commitments.Today, this kind of optimisation is frequently being done with the aid of&nbsp;<a href="https://www.rowse.co.uk/blog/post/an-introduction-to-digital-twins" target="_blank" title="An Introduction To Digital Twins">digital twins</a>. These are valuable in both design and planning environments, and in speeding up construction and trialling of equipment. They can also be valuable in providingreal-time simulations of your supply chain. The constant input of data allows the model to reflect the actual conditions existing in your supply chain as they evolve. This provides you with a collaborative environment where you can test out and verify different hypotheses using AI-powered models.Many manufacturing and logistics enterprises are already deploying some kind of supply chain digital twin to assist their planning. This means that the biggest obstacle facing many companies trying to optimise their network is that of scaling. Digital twin technology has many different use cases, and will require some nominal investment, but as industry becomes more accustomed to AI, it&#39;ll be worth it.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1659945064475-1659945064475.png" alt="How To Use AI/ML To Optimise Manufacturing Costs" class="fr-fic fr-dii"><h2>Detailed Scheduling</h2>When you&#39;re on the real factory floor, planning and forecasting comes down to the detailed scheduling of production runs. Human beings previously planned their schedules simply by using their eyes and brain to assess the situation. You might have moved on to incorporate a digital planning solution, using simple software that doesn&rsquo;t have integrated data analytics capabilities. These were all very well in their day, but today&#39;s production programs incorporate many more constraints, and you might find it much harder to create an optimal route forward.Taking into account the global disruption caused by COVID-19, climate change and political unrest, it&#39;s much harder to plan ahead for every eventuality. Production constraints include such things as raw material requirements and availability, machine and manpower capacity, and customer requirements. In today&#39;s climate, when technologies such as CAD and&nbsp;<a href="https://www.rowse.co.uk/blog/post/3d-printing-in-manufacturing" target="_blank" title="3D Printing In Manufacturing">3D printing</a> allow for a much greater degree of customisation, you also need to consider what parts and materials are necessary not just for individual products, but for individualised items.All of this creates a complex array of potentialities that can be difficult to evaluate. This is where AI has the upper hand, as it&#39;s designed to deal with precisely this kind of complexity.If you adopt an AI-powered planning solution for your operation, detailed scheduling will be a breeze, even in a job shop environment. You&#39;ll be able to structure any type or volume of workload on an unimaginably minute level of detail. It will all be accomplished in much less time than ever before, enabling much greater agility in your operation. You&#39;ll respond more flexibly to both actual and projected orders, reduce wastage and downtime, and improve your on-time performance (OTP) rate.This OTP rate is amongst the most commonly measured and critical supply chain KPIs when it comes to eCommerce, and your customers will benefit greatly from the improvements that AI can deliver.<h2>Forecasting Accuracy</h2>Once you move into the more abstract realms of an AI-driven technology ecosystem, you&#39;re looking at financial planning and accounting. This involves forecasting future demand so as to improve your sales and operations process. You can use AI to analyse historical data and combine it with current market trends. It can then deploy algorithms for more predictive modelling, so that your forecasts will be much more accurate.You&#39;ll also be able to manage the costs associated with shortages and surpluses, by matching demand and capacity more effectively. In the long run, you&#39;ll have a better prediction of what you&#39;ll need in your supply chain, thus improving your margins and boosting your top line.The COVID-19 pandemic has caused significant delays in obtaining and shipping parts and finished goods, with the automotive industry being visibly impacted. In addition, the supply chain is still suffering from one-off impacts such as the 2021 blockage of the Suez Canal. This has influenced the global manufacturing sector to accelerate adoption of AI/ML capabilities, so that companies are better able to model such previously inconceivable scenarios. Recent news of another grounding by a container ship from the same stable as the Ever Given may well underscore manufacturers&#39; concerns.Increased forecasting accuracy will also enable manufacturing companies greater precision in purchasing, and thus help to reduce storage costs. The lag time that pushes up costs should be eradicated, and the necessity for expedited shipping reduced &ndash; along with penalties imposed by customers for late delivery. The AI/ML enhancement of forecasting will affect other accounting procedures, such as assessments of asset impairment recoverability, and obsolete or excess inventory reserves. Risk assessments will also be affected as these underlying accounting processes evolve.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1659945064250-1659945064250.png" alt="How To Use AI/ML To Optimise Manufacturing Costs" class="fr-fic fr-dii"><h2>Optimisation Through AI And ML</h2>AI, ML and big data analysis can offer substantial&nbsp;<a href="https://www.rowse.co.uk/blog/post/benefits-of-industry-4" target="_blank" title="Benefits Of Industry 4.0">business benefits</a> to industry. The average modern automated factory produces enormous caches of data every year, and these technologies are information-driven. AI and ML algorithms are thriving on analysis of this data, in order to optimise efficiency and save costs. Many manufacturers are beginning to leverage analytics technology to enhance their existing capacity, improving performance and reducing downtime.Digital transformation will spread exponentially, as the adoption of AI technologies reveals further opportunities for industry to adapt its cost structures. There are many other areas that will be influenced, including changes in jobs and staffing. Manufacturers should study these carefully to find the best way of optimising their efficiency and costs.<hr>This article was originally published at:&nbsp;<a href="https://www.rowse.co.uk/blog/post/how-to-use-ai-ml-to-optimise-manufacturing-costs" id="isPasted" rel="noopener noreferrer" target="_blank">How To Use AI/ML To Optimise Manufacturing Costs</a>

How To Use AI/ML To Optimise Manufacturing Costs

Before the pandemic, research suggested that adoption of AI (artificial intelligence) and ML (machine learning) in business was increasing at a rate of about 25% per year.

How To Use AI ML To Optimise Manufacturing Costs

It would be awesome to have robots that can help us do important things, like aid disaster recovery efforts or monitor the environment. In the case of quadrupeds, robots that walk on four legs, their mobility requires many software components to work together seamlessly. It&rsquo;s a drag when all your time is spent on developing lower-level infrastructure instead of focusing on high-level behaviors. <a href="https://engineering.cmu.edu/directory/bios/johnson-aaron.html" id="isPasted" target="_blank">Aaron Johnson</a>&rsquo;s <a href="https://www.cmu.edu/me/robomechanicslab/" rel="noopener" target="_blank">Robomechanics Lab</a> experienced these frustrations firsthand. His team often had to rely on simple models for doing research because existing software solutions were not open sourced, did not provide a modular framework, and lacked end-to-end functionality.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjU5NzE4NTkzNzMxLTA3MjYtZW0tMS1xdWFkcnVwZWQucG5nIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==" style="width: 766px;" class="fr-fic fr-dib">A four-legged robot using Quad-SDK software is shown jumping. Source:&nbsp;Carnegie Mellon University/Robomechanics Lab&nbsp;An example of innovation born out of necessity, Johnson, an associate professor of <a href="https://www.meche.engineering.cmu.edu/index.html" id="isPasted" rel="noopener" target="_blank">mechanical engineering</a>, and his team designed their own locomotion software stack, <a href="https://github.com/robomechanics/quad-sdk" rel="noopener" target="_blank">Quad-SDK</a>, a full-stack framework for agile quadrupedal locomotion. Many graduate students have contributed to the development of the software across a two-year period, but it was time well spent. Their design can simplify the development process for roboticists everywhere because it uses an open-source license, meaning the software can be used and modified as the user wishes. Quad-SDK comes ready to use, so researchers don&rsquo;t have to worry about implementing the necessary tools and infrastructure; instead, they can get right to work on exciting behaviors and applications.&nbsp;<iframe src="https://www.youtube.com/embed/kSXKjTxKpuA?&wmode=opaque&rel=0&enablejsapi=1&origin=https://www.wevolver.com" frameborder="0" allowfullscreen="" class="fr-draggable" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true" style="width: 767px; height: 430px;"></iframe>Unlike other options, Quad-SDK is also compatible with the Robot Operating System (ROS). ROS is a middleware, somewhere between hardware and software, that lets different parts of a system talk to each other. For example, if a robot senses an obstacle in its path and needs to transfer information from its perception module to its decision-making module, ROS is what enables that communication. Imagine an app that doesn&rsquo;t use iOS or Android&mdash;it&rsquo;s much simpler when everything works together.Locomotion is a layered problem, as team member and Ph.D. student Ardalan Tajbakhsh describes it. &ldquo;In order to do anything meaningful on a robot, you need to have many components working together seamlessly.&rdquo; Quad-SDK provides a framework for robotics researchers and developers to focus their efforts on the core algorithms instead of the software tooling and infrastructure.<blockquote>In order to do anything meaningful on a robot, you need to have many components working together seamlessly. - Ardalan Tajbakhsh, Ph.D. student, Mechanical Engineering&nbsp;</blockquote>Other software packages are very good at solving one component, like motion planning, but it&rsquo;s critical to have end-to-end frameworks that provide the necessary algorithms, tools, and infrastructure for performing high-quality robotics research. Quad-SDK is full stack, meaning it contains every level of the hierarchy that affects quadruped locomotion, starting with global planning. This layer sits at the top of the stack; Tajbakhsh compares it to Google Maps, because it chooses where the robot should roughly go to reach its destination. The next layer, local planner, is what decides the specifics of the route, like where the robot should place its feet. This culminates in a third layer, the robot driver, which sends commands to the quadruped&rsquo;s joints to execute the desired motion.In the weeks since Quad-SDK <a href="https://leggedrobots.org/contributions.html" rel="noopener" target="_blank">won a Best Paper Award</a> at the Workshop on Legged Robots at the 2022 IEEE International Conference on Robotics and Automation (ICRA), people have already requested to use it, a clear indicator of how useful its solutions are and how the open-source format invites collaboration within the robotics community. Figuratively and literally, Quad-SDK can go many places.

machine learning

Latest Posts