Mediaboss Marketing

Thanks to machine learning algorithms, short pieces of DNA floating in the bloodstream of cancer patients can help doctors diagnose specific types of cancer and choose the most effective treatment for a patient.

The new analysis technique, created by University of WisconsinMadison researchersandpublished recently in Annals of Oncology, is compatible with liquid biopsy testing equipment already approved in the United States and in use in cancer clinics. This could speed the new methods path to helping patients.

Liquid biopsies rely on simple blood draws instead of taking a piece of cancerous tissue from a tumor with a needle.

Marina Sharifi

Liquid biopsies are much less invasive than a tissue biopsy which may even be impossible to do in some cases, depending on where a patients tumor is, saysMarina Sharifi, a professor of medicine and oncologist in UWMadisons School of Medicine and Public Health. Its much easier to do them multiple times over the course of a patients disease to monitor the status of cancer and its response to treatment.

Cancerous tumors shed genetic material, called cell-free DNA, into the bloodstream as they grow. But not all parts of a cancer cells DNA are likely to tumble away. Cells store some of their DNA by coiling it up in protective balls called histones. They unwrap sections to access parts of the genetic code as needed.

Kyle Helzer, a UWMadison bioinformatics scientist, says that parts of the DNA containing the genes that cancer cells use often are uncoiled more frequently and thus are more likely to fragment.

Were exploiting that larger distribution of those regions among cell-free DNA to identify cancer types, adds Helzer, who is also a co-lead author of the study along with Sharifi and scientist Jamie Sperger.

Shuang Zhao

The research team, led by UWMadison senior authorsShuang (George) Zhao, professor of human oncology, andJoshua Lang, professor of medicine, used DNA fragments found in blood samples from a past study of nearly 200 patients (some with, some without cancer), and new samples collected from more than 300 patients treated for breast, lung, prostate or bladder cancers at UWMadison and other research hospitals in the Big Ten Cancer Research Consortium.

The scientists divided each group of samples into two. One portion was used to train a machine-learning algorithm to identify patterns among the fragments of cell-free DNA, relatively unique fingerprints specific to different types of cancers. They used the other portion to test the trained algorithm. The algorithm topped 80 percent accuracy translating the results of a liquid biopsy into both a cancer diagnosis and the specific types of cancer afflicting a patient.

In addition, the machine learning approach was able to tell apart two subtypes of prostate cancer: the most common version, adenocarcinoma, and a swift-progressing variant called neuroendocrine prostate cancer (NEPC) that is resistant to standard treatment approaches. Because NEPC is often difficult to distinguish from adenocarcinoma, but requires aggressive action, it puts oncologists like Lang and Sharifi in a bind.

Joshua Lang

Currently, the only way to diagnose NEPC is via a needle biopsy of a tumor site, and it can be difficult to get a conclusive answer from this approach, even if we have a high clinical suspicion for NEPC, Sharifi says.

Liquid biopsies have advantages, Sperger adds, in that you dont have to know which tumor site to biopsy at, and it is much easier for the patient to get a standard blood draw.

The blood samples were processed using cell-free DNA sequencing technology marketed by Iowa-based Integrated DNA Technologies. Using standard panels like those currently in the clinic is a departure one that can reduce the time and cost of testing from other methods of fragmentomic analysis of cancer DNA in blood samples.

Most commercial panels have been developed around the most important cancer genes that indicate certain drugs for treatment, and they sequence those select genes, says Zhao. What weve shown is that we can use those same panels and same targeted genes to look at the fragmentomics of the cell-free DNA in a blood sample and identify the type of cancer a patient has.

The UW Carbone Cancer Centers Circulating Biomarker Core and Biospecimen Disease-Oriented Team contributed to the collection of the studys hundreds of patient samples.

This research was funded in part by grants from the National Institutes of Health (DP2 OD030734, 1UH2CA260389 and R01CA247479) and the Department of Defense (PC190039, PC200334 and PC180469.)

Written by Chris Barncard

Link to original story: https://news.wisc.edu/algorithmic-blood-test-analysis-will-ease-diagnosis-of-cancer-types-guide-treatment/

The rest is here:
UW-Madison: Cancer diagnosis and treatment could get a boost ... - University of Wisconsin System

UC Merced Computer Science and Engineering Professor Xiaoyi Luis leading a collaboration that secureda $4.35 million grant from the Department of Energy (DOE) to improve federated machine learning systems.

Lu is partnering with the University of Iowa and Argonne National Laboratory near Chicago to improve the understanding of scalable, federated, privacy-preserving machine learning. This project is one of five initiatives centered on distributed resilient systems in science that have collectively received $40 million in funding from the DOE.

"Scientific research is getting more complex and will need next-generation workflows as we move forward with larger data sets and new tools spread across the U.S.," Ceren Susut, DOE acting Associate Director of Science for Advanced Scientific Computing Research, said in a news release announcing the awards. "This program will explore how science can be conducted in this new environment - where tools and data are in multiple places but must be integrated in a high-performance fashion."

According to his abstract, Lu's proposal "aims to address the critical need for a scalable and resilient Federated Learning simulation and modeling system in the context of edge computing-related scientific research and exploration."

Federated Learning embodies a decentralized approach to training machine learning models, placing a strong emphasis on enhancing data privacy. In contrast to the traditional method that requires data to be transferred from client devices to global servers, Federated Learning harnesses raw data residing on edge devices to facilitate local model training. These edge devices, responsible for connecting various devices and facilitating network traffic between them, assume a pivotal role in this process.

"Federated learning is becoming an essential technique for machine learning on edge devices as the sheer amount of raw data generated by these devices requires real-time, effective data processing at the edge device ends," Lu wrote in his abstract. "The processed data carrying intelligent information must be encrypted for privacy protection, making federated learning the best solution for building a well-trained model across decentralized smart edge devices with secure and efficient data-sharing policies."

Lu and his partners propose a scalable and resilient federated learning simulation and modeling system. This system will empower users to harness privacy-preserving algorithms, introduce novel algorithms, and simulate as well as deploy a wide range of federated learning algorithms with privacy-preserving techniques.

"The proposed system brings forth substantial advantages for researchers and developers engaged in real-world federated learning systems," Lu explained. "It furnishes them with a valuable platform for conducting proof-of-concept implementations and performance validation, which are essential prerequisites before deploying and testing their machine learning models in real-world contexts. Additionally, the proposed system is poised to make a significant scientific impact on DOE-mission-based applications, including scientific machine learning and critical infrastructure, where concerns regarding data privacy hold significant weight."

See the original post here:
Department of Energy Grant will Fund EECS Professor Lu's ... - University of California, Merced

A machine learning model created at CSU has improved forecasters confidence in storm predictions and is now used daily by the National Weather Services Storm Prediction Center and Weather Prediction Center.

The model, developed in the Department of Atmospheric Science by a team led by Schumacher, is capable of accurately predicting excessive rainfall, hail and tornadoes four to eight days in advance. The model is called CSU-MLP for Colorado State University-Machine Learning Probabilities.

Schumachers team worked with NWS forecasters over six years to test and refine the model for their purposes. The CSU code is now running on the Storm Prediction Centers and Weather Prediction Centers operational computer systems, helping forecasters predict hazardous weather, so people in harms way have enough lead time to prepare.

The atmospheric scientists trained the model on historical records of severe weather and NOAA reforecasts, retrospective forecasts run with todays improved numerical models.

Team member Allie Mazurek, a Ph.D. student, is working on explainable AI for the CSU-MLP forecasts. Shes trying to figure out which atmospheric data inputs are most important to the models predictions, so the model will be more transparent for forecasters.

These new tools that use AI for weather prediction are developing quickly and showing some really promising and exciting results, Schumacher said. But they also have limitations, just like traditional weather prediction models and human forecasters have strengths and limitations. The best way to advance the field and improve forecasts will be to take advantage of each of their strengths: the AI for what its good at, which is identifying patterns in massive datasets; numerical weather prediction models for being grounded in the physics; and humans for synthesizing, understanding and communicating.

Schumacher discusses the promise and limitations of AI for weather prediction in more detail in this piece in The Conversation, co-authored by Aaron Hill, a former CSU research scientist who is now a faculty member at the University of Oklahoma.

Read more from the original source:
Can AI help in climate change? CSU researchers have the answer. - Source

The Internet of Things (IoT), a network of interconnected devices equipped with sensors and software, has revolutionised how we interact with the world around us, empowering us to collect and analyse data like never before.

As technology advances and becomes more accessible, more objects are equipped with connectivity and sensor capabilities, making them part of the IoT ecosystem. The number of active IoT systems is expected to reach 29.7 billion by 2027, marking a significant surge from the 3.6 billion devices recorded in 2015. This exponential growth requires a tremendous demand for solutions to mitigate the safety and computational challenges of IoT applications. In particular, industrial IoT, automotive, and smart homes are three main areas with specific requirements, but they share a common need for efficient IoT systems to enable optimal functionality and performance.

Increasing the efficiency of IoT systems and unlocking their potential can be achieved through Artificial Intelligence (AI), creating AIoT architectures. By utilising sophisticated algorithms and Machine Learning techniques, AI empowers IoT systems to make intelligent decisions, process vast amounts of data, and extract valuable insights. For instance, this integration drives operational optimisation in industrial IoT, facilitates advanced autonomous vehicles, and offers intelligent energy management and personalised experiences in smart homes.

Among the different AI algorithms, Deep Learning that leverages artificial neural networks is very appropriate for IoT systems for several reasons. One of the primary reasons is its ability to learn and extract features automatically from raw sensor data. This is particularly valuable in IoT applications where the data can be unstructured, noisy, or have complex relationships. Additionally, Deep Learning enables IoT applications to handle real-time and streaming data efficiently. This ability allows for continuous analysis and decision-making, which is crucial in time-sensitive applications such as real-time monitoring, predictive maintenance, or autonomous control systems.

Despite the numerous advantages of Deep Learning for IoT systems, its implementation has inherent challenges, such as efficiency and safety, that must be addressed to fully leverage its potential. The Very Efficient Deep Learning in IoT (VEDLIoT) project aims to solve these challenges.

A high-level overview of the different VEDLIoT components is given in Fig. 1. IoT is integrated with Deep Learning by the VEDLIoT project to accelerate applications and optimise the energy efficiency of IoT. VEDLIoT achieves these objectives through the utilisation of several key components:

VEDLIoT concentrates on some use cases, such as demand-oriented interaction methods in smart homes (see Fig. 2), industrial IoT applications like Motor Condition Classification and Arc Detection, and the Pedestrian Automatic Emergency Braking (PAEB) system in the automotive sector (see Fig. 3). VEDLIoT systematically optimises such use cases through a bottom-up approach by employing requirement engineering and verification techniques, as shown in Fig. 1. The project combines expert-level knowledge from diverse domains to create a robust middleware that facilitates development through testing, benchmarking, and deployment frameworks, ultimately ensuring the optimisation and effectiveness of Deep Learning algorithms within IoT systems. In the following sections, we briefly present each component of the VEDLIoT project.

Various accelerators are available for a wide range of applications, from small embedded systems with power budgets in the milliwatt range to high-power cloud platforms. These accelerators are categorised into three main groups based on their peak performance values, as shown in Fig. 4.

The first group is the ultra-low power category (< 3 W), which consists of energy-efficient microcontroller-style cores combined with compact accelerators for specific Deep Learning functions. These accelerators are designed for IoT applications and offer simple interfaces for easy integration. Some accelerators in this category provide camera or audio interfaces, enabling efficient vision or sound processing tasks. They may offer a generic USB interface, allowing them to function as accelerator devices attached to a host processor. These ultra-low power accelerators are ideal for IoT applications where energy efficiency and compactness are key considerations, providing optimised performance for Deep Learning tasks without excessive power.

The VEDLIoT use case of predictive maintenance is a good example and makes use of an ultra-low power accelerator. One of the most important design criteria is low power consumption, as it is a battery-powered small box that can externally be installed on any electric motor and should monitor the electronic motor for at least three years without a battery change.

The next category is the low-power group (3 W to 35 W), which targets a broad range of automation and automotive applications. These accelerators feature high-speed interfaces for external memories and peripherals and efficient communication with other processing devices or host systems such as PCIe. They support modular and microserver-based approaches and provide compatibility with various platforms. Additionally, many accelerators in this category incorporate powerful application processors capable of running full Linux operating systems, allowing for flexible software development and integration. Some devices in this category include dedicated application-specific integrated circuits (ASICs), while others feature NVIDIAs embedded graphics processing units (GPUs). These accelerators balance power efficiency and processing capabilities, making them well-suited for various compute-intensive tasks in the automation and automotive domains.

The high-performance category (> 35 W) of accelerators is designed for demanding inference and training scenarios in edge and cloud servers. These accelerators offer exceptional processing power, making them suitable for computationally-intensive tasks. They are commonly deployed as PCIe extension cards and provide high-speed interfaces for efficient data transfer. The devices in this category have high thermal design powers (TDPs), indicating their ability to handle significant workloads. These accelerators include dedicated ASICs, known for their specialised performance in Deep Learning tasks. They deliver accelerated processing capabilities, enabling faster inference and training times. Some consumer-class GPUs may also be included in benchmarking comparisons to provide a broader perspective.

Selecting the proper accelerator from the abovementioned wide range of available options is not straightforward. However, VEDLIoT takes on this crucial responsibility by conducting thorough assessments and evaluations of various architectures, including GPUs, field-programmable gate arrays (FPGAs), and ASICs. The project carefully examines these accelerators performances and energy consumptions to ensure their suitability for specific use cases. By leveraging its expertise and comprehensive evaluation process, VEDLIoT guides the selection of Deep Learning accelerators within the project and in the broader landscape of IoT and Deep Learning applications.

Trained Deep Learning models have redundancy that can sometimes be compressed to 49 times their original size, with negligible accuracy loss. Although many works are related to such compression, most results show theoretical speed-ups that only sometimes translate into more efficient hardware execution since they do not consider the target hardware. On the other hand, the process of deploying Deep Learning models on edge devices involves several steps, such as training, optimisation, compilation, and runtime. Although various frameworks are available for these steps, their interoperability can vary, resulting in different outcomes and performance levels. VEDLIoT addresses these challenges through hardware-aware model optimisation using ONNX, an open format for representing Machine Learning models, ensuring compatibility with the current open ecosystem. Additionally, Renode, an open-source simulation framework, serves as a functional simulator for complex heterogeneous systems, allowing for the simulation of complete System-on-Chips (SoCs) and the execution of the same software used on hardware.

Furthermore, VEDLIoT uses the EmbeDL toolkit to optimise Deep Learning models. The EmbeDL toolkit offers comprehensive tools and techniques to optimise Deep Learning models for efficient deployment on resource-constrained devices. By considering hardware-specific constraints and characteristics, the toolkit enables developers to compress, quantise, prune, and optimise models while minimising resource utilisation and maintaining high inference accuracy. EmbeDL focuses on hardware-aware optimisation and ensures that Deep Learning models can be effectively deployed on edge devices and IoT devices, unlocking the potential for intelligent applications in various domains. With EmbeDL, developers can achieve superior performance, faster inference, and improved energy efficiency, making it an essential resource for those seeking to maximise the potential of Deep Learning in real-world applications.

Since VEDLIoT aims to combine Deep Learning with IoT systems, ensuring security and safety becomes crucial. In order to emphasise these aspects in its core, the project leverages trusted execution environments (TEEs), such as Intel SGX and ARM TrustZone, along with open-source runtimes like WebAssembly. TEEs provide secure environments that isolate critical software components and protect against unauthorised access and tampering. By using WebAssembly, VEDLIoT offers a common environment for execution throughout the entire continuum, from IoT, through the edge and into the cloud.

In the context of TEEs, VEDLIoT introduces Twine and WaTZ as trusted runtimes for Intels SGX and ARMs TrustZone, respectively. These runtimes simplify software creation within secure environments by leveraging WebAssembly and its modular interface. This integration bridges the gap between trusted execution environments and AIoT, helping to seamlessly integrate Deep Learning frameworks. Within TEEs using WebAssembly, VEDLIoT achieves hardware-independent robust protection against malicious interference, preserving the confidentiality of both data and Deep Learning models. This integration highlights VEDLIoTs commitment to securing critical software components, enabling secure development, and facilitating privacy-enhanced AIoT applications in cloud-edge environments.

Additionally, VEDLIoT employs a specialised architectural framework, as shown in Fig. 5, that helps to define, synchronise and co-ordinate requirements and specifications of AI components and traditional IoT system elements. This framework consists of various architectural views that address the systems specific design concerns and quality aspects, including security and ethical considerations. By using these architecture views as templates and filling them out, correspondences and dependencies can be identified between the quality-defining architecture views and other design decisions, such as AI model construction, data selection, and communication architecture. This holistic approach ensures that security and ethical aspects are seamlessly integrated into the overall system design, reinforcing VEDLIoTs commitment to robustness and addressing emerging challenges in AI-enabled IoT systems.

Traditional hardware platforms support only homogeneous IoT systems. However, RECS, an AI-enabled microserver hardware platform, allows for the seamless integration of diverse technologies. Thus, it enables fine-tuning of the platform towards specific applications, providing a comprehensive cloud-to-edge platform. All RECS variants share the same design paradigm to be a densely-coupled, highly-integrated communication infrastructure. For the varying RECS variants, different microserver sizes are used, from credit card size to tablet size. This allows customers to choose the best variant for each use case and scenario. Fig. 6 gives an overview of the RECS variants.

The three different RECS platforms are suitable for cloud/data centre (RECS|Box), edge (t.RECS) and IoT usage (u.RECS). All RECS servers use industry-standard microservers, which are exchangeable and allow for use of the latest technology just by changing a microserver. Hardware providers of these microservers offer a wide spectrum of different computing architectures like Intel, AMD and ARM CPUs, FPGAs and combinations of a CPU with an embedded GPU or AI accelerator.

VEDLIoT addresses the challenge of bringing Deep Learning to IoT devices with limited computing performance and low-power budgets. The VEDLIoT AIoT hardware platform provides optimised hardware components and additional accelerators for IoT applications covering the entire spectrum, from embedded via edge to the cloud. On the other hand, a powerful middleware is employed to ease the programming, testing, and deployment of neural networks in heterogeneous hardware. New methodologies for requirement engineering, coupled with safety and security concepts, are incorporated throughout the complete framework. The concepts are tested and driven by challenging use cases in key industry sectors like automotive, automation, and smart homes.

Please note, this article will also appear in the fifteenthedition of ourquarterly publication.

See the rest here:
Unlocking the potential of IoT systems: The role of Deep Learning ... - Innovation News Network

Modern deep-learning algorithms are now focused on problem environments where training occurs just once on a sizable data collection, never againall of the early triumphs of deep learning in voice recognition and picture classification employed such train-once settings. Replay buffers and batching were later added to deep understanding when applied to reinforcement learning, making it extremely close to a train-once setting. A large batch of data was also used to train recent deep learning systems like GPT-3 and DallE. The most popular approach in these situations has been to gather data continuously and then occasionally prepare a new network from scratch in a training configuration. Of course, in many applications, the data distribution varies over time, and training must continue in some manner. Modern deep-learning techniques were developed with the train-once setting in mind.

In contrast, the perpetual learning problem setting focuses on continuously learning from fresh data. The ongoing learning option is ideal for issues where the learning system must deal with a dynamic data stream. For instance, think of a robot that has to find its way around a house. The robot would have to be retrained from scratch or run the danger of being rendered useless every time the houses layout changed if the train-once setting was used. It would be necessary to retrain from scratch if the design changed regularly. On the other hand, the robot might easily learn from the new information and continuously adjust to the changes in the house under the ongoing learning scenario. The importance of lifelong learning has grown in recent years, and more specialized conferences are being held to address it, such as the Conference on Life-long Learning Agents (CoLLAS).

They emphasize the environment of ongoing learning in their essay. When exposed to fresh data, deep learning systems frequently lose most of what they have previously learned, a condition known as catastrophic forgetting. In other words, deep learning techniques do not retain stability in ongoing learning issues. In the late 1900s, early neural networks were the first to demonstrate this behavior. Catastrophic forgetting has recently gotten fresh interest due to the development of deep learning since several articles have been written about preserving stability in deep continuous learning.

The capacity to continue learning from fresh material is distinct from catastrophic forgetting and perhaps more essential to continuous learning. They call this capacity plasticity.Continuous learning systems must maintain plasticity because it enables them to adjust to changes in their data streams. If their data stream changes, continuously learning systems that lose flexibility may become worthless. They emphasize the problem of flexibility loss in their essay. These studies employed a configuration in which the network was first shown a collection of instances for a predetermined number of epochs, after which the training set was enlarged with new examples, and the training cycle repeated for an extra number of epochs. After accounting for the number of epochs, they discovered that the error for the cases in the first training set was lower than for the later-added examples. These publications offered proof that the loss of flexibility caused by deep learning and the backpropagation algorithm upon which it is based is a common occurrence.

New outputs, known as heads, were added to the network in its configuration when a new job was offered, and the number of outputs increased as more tasks were encountered. Thus, the effects of interference from old heads were mixed up with the consequences of plasticity loss. According to Chaudhry et al., the loss of plasticity was modest when old heads were taken out at the beginning of a new task, indicating that the major cause of the loss of plasticity they saw was interference from old heads. The fact that previously researchers only employed ten challenges prevented them from measuring the loss of plasticity that occurs when deep learning techniques are presented with a lengthy list of tasks.

Although the findings in these publications suggest that deep learning systems have lost some of their essential adaptability, no one has yet shown that continuous learning has lost plasticity. In the reinforcement learning field, where recent works have demonstrated a significant loss of plasticity, there is more evidence for the loss of plasticity in contemporary deep learning. By demonstrating that early learning in reinforcement learning issues can have a negative impact on later learning, Nishikin et al. coined the term primacy bias.

Given that reinforcement learning is fundamentally continuous as a consequence of changes in the policy, this result may be attributable to deep learning networks losing their flexibility in circumstances where learning is ongoing. Additionally, Lyle et al. demonstrated that some deep reinforcement learning agents may eventually lose their capacity to pick up new skills. These are significant data points, but because of the intricacy of contemporary deep reinforcement learning, it isnt easy to make any firm conclusions. These studies show that deep learning systems lose flexibility but fall short of providing a complete explanation of the phenomenon. These studies include those from the psychology literature around the turn of the century and more contemporary ones in machine learning and reinforcement learning. In this study, researchers from the Department of Computing Science, University of Alberta, and CIFAR AI Chair, Alberta Machine Intelligence Institute provide a more conclusive response to plasticity loss in contemporary deep learning.

They demonstrate that persistent supervised learning issues cause deep learning approaches to lose plasticity and that this plasticity loss can be severe. In a continuous supervised learning problem using the ImageNet dataset and including hundreds of learning trials, they first show that deep learning suffers from loss of plasticity. The complexity and related confusion that always develop in reinforcement learning are eliminated when supervised learning tasks are used instead. We can also determine the complete amount of the loss of plasticity thanks to the hundreds of tasks that we have. They next prove the universality of deep learnings lack of flexibility over a wide variety of hyperparameters, optimizers, network sizes, and activation functions using two computationally less expensive problems (a variation of MNIST and the slowly changing regression problem). They want a deeper grasp of its origins after demonstrating the severity and generality of loss of flexibility in deep learning.

Check out thePaper.All Credit For This Research Goes To the Researchers on This Project. Also,dont forget to joinour 29k+ ML SubReddit,40k+ Facebook Community,Discord Channel,andEmail Newsletter, where we share the latest AI research news, cool AI projects, and more.

If you like our work, you will love our newsletter..

Aneesh Tickoo is a consulting intern at MarktechPost. He is currently pursuing his undergraduate degree in Data Science and Artificial Intelligence from the Indian Institute of Technology(IIT), Bhilai. He spends most of his time working on projects aimed at harnessing the power of machine learning. His research interest is image processing and is passionate about building solutions around it. He loves to connect with people and collaborate on interesting projects.

Read more:
This AI Research Addresses the Problem of 'Loss of Plasticity' in ... - MarkTechPost