Machine Learning
Machine learning (ML) is by no means new. It’s been around for decades. But thanks to big data and more-powerful computers, it has evolved into an amazing tool that has helped to advance—and often speed up—scientific research. In machine learning, computer systems automatically learn from experience without being explicitly programmed. A computer program analyzes data to look for patterns; determines the complex statistical structures that identify specific features; and then finds those same features in new data.
ML now supports much of the technology around us: smartphone cameras recognizing faces; online translating and captioning; credit-card fraud detecting; personalized online marketing; and providing newsfeeds. In the NIH intramural program, machine-learning techniques are being used in several research areas: image processing; natural-language processing; genomics; drug discovery; and studies of disease prediction, detection, and progression.
In work described in the Journal of the National Cancer Institute, researchers used a high-performance computer to analyze thousands of cervical photographs, obtained more than 20 years ago from volunteers in a cancer screening study. The computer learned to recognize specific patterns associated with pre-cancerous and cancerous changes of the cervix, and that information was used to develop an algorithm for reliably detecting such changes in the collection of images. In fact, the AI-generated algorithm outperformed human expert reviewers and all standard screening tests in detecting pre-cancerous changes.
At Fermi National Accelerator Laboratory the Inter-experimental Machine Learning Working Group brings together a community of analyzers of HEP data who use and develop machine learning (ML) algorithms to solve physics problems. They share tools and applications, provide training and discuss challenges related to the use of ML tools in the HEP community. They hold regular meetings focused on tool development, knowledge transfer and common solutions to known challenges. Experts from multiple experiments provide feedback and encourage collaboration among members in order to promote the use of ML tools in the community for problems for which they have been shown to be useful. The sister group, the Inter-Experimental LHC Machine Learning (IML) working group, is focused on building a community of researchers in machine learning in particle physics that brings together interested people from different experiments and external machine learning experts in academia and industry.
ML is a type of artificial intelligence (AI) that enables computers to learn automatically without being explicitly programmed. It focuses on the development of computer programs that can access data and use it to learn for themselves. The learning process begins with making observations in data, identifying patterns and making better decisions in the future based on the examples that researchers provide. In its more advanced form, machine learning could enable computers to look for new physics without being explicitly told what to look for.
The past several years have seen a revolution in ML. New hardware, especially Graphics Processing Units (GPUs), and the algorithms associated with deep learning have enabled computers to surpass humans in certain pattern recognition exercises for the first time. Deep learning now dominates research in AI and has found wide application across many problem domains. These techniques have the potential to greatly amplify the ability to do science and, indeed, have already begun to impact experiments at Fermilab. In addition, these algorithms are broadly applicable to various nonlinear optimization problems, including (for an institution like Fermilab) accelerator operations.
In the Scientific Computing Division, they work to empower the research program to further the laboratory’s mission by providing centralized access to expertise and resources. Their goal is to enable scientists at the lab to deploy machine learning solutions through consulting and training. They are a resource for groups just getting started in ML and also for experts who understand the techniques and just need help understanding the infrastructure available to them and build bridges to experts outside of the lab through seminar series and by representing the Fermilab community at machine learning conferences and workshops.
The current generation of ML systems would not have been possible without significant computing advances made over the past few decades. The development of the graphics-processing unit (GPU) was critical to the advancement of ML as it provided new levels of compute power needed for ML systems to process and train on large data sets. As the field of artificial intelligence looks towards advancing beyond today’s ML capabilities, pushing into the realms of learning in real-time, new levels of computing are required. Highly specialized Application-Specific Integrated Circuits (ASICs) show promise in meeting the physical size, weight, and power (SWaP) requirements of advanced ML applications, such as autonomous systems and 5G. However, the high cost of design and implementation has made the development of ML-specific ASICs impractical for all but the highest volume applications.
“A critical challenge in computing is the creation of processors that can proactively interpret and learn from data in real-time, apply previous knowledge to solve unfamiliar problems, and operate with the energy efficiency of the human brain,” said Andreas Olofsson, a program manager in DARPA’s Microsystems Technology Office (MTO). “Competing challenges of low-SWaP, low-latency, and adaptability require the development of novel algorithms and circuits specifically for real-time machine learning. What’s needed is the rapid development of energy efficient hardware and ML architectures that can learn from a continuous stream of new data in real time.”
DARPA’s Real Time Machine Learning (RTML) program seeks to reduce the design costs associated with developing ASICs tailored for emerging ML applications by developing a means of automatically generating novel chip designs based on ML frameworks. The goal of the RTML program is to create a compiler – or software platform – that can ingest ML frameworks like TensorFlow and Pytorch and, based on the objectives of the specific ML algorithms or systems, generate hardware design configurations and standard Verilog code optimized for the specific need. Throughout the lifetime of the program, RTML will explore the compiler’s capabilities across two critical, high-bandwidth application areas: 5G networks and image processing.
“Machine learning experts are proficient in developing algorithms but have little to no knowledge of chip design. Conversely, chip designers are not equipped with the expertise needed to inform the design of ML-specific ASICs. RTML seeks to merge these unique areas of expertise, making the process of designing ultra-specialized ASICs more efficient and cost-effective,” said Olofsson.
Based on the application space’s anticipated agility and efficiency, the RTML compiler provides an ideal platform for prototyping and testing fundamental ML research ideas that require novel chip designs. As such, DARPA plans to collaborate with the National Science Foundation (NSF) on this effort. NSF is pursuing its own Real Time Machine Learning program focused on developing novel ML paradigms and architectures that can support real-time inference and rapid learning. After the first phase of the DARPA RTML program, the agency plans to make its compiler available to NSF researchers to provide a platform for evaluating their proposed ML algorithms and architectures. During the second phase of the program, DARPA researchers will have an opportunity to evaluate the compiler’s performance and capabilities using the results generated by NSF. The overall expectation of the DARPA-NSF partnership is to lay the foundation for next-generation co-design of RTML algorithms and hardware.
RTML is part of the second phase of DARPA’s Electronics Resurgence Initiative (ERI) – a five-year, upwards of $1.5 billion investment in the future of domestic, U.S. government, and defense electronics systems. As a part of ERI Phase II, DARPA is supporting domestic manufacturing options and enabling the development of differentiated capabilities for diverse needs. RTML is helping to fulfill this mission by creating a means of expeditiously and cost-effectively generating novel chip designs to support emerging ML applications.
Although Artificial intelligence (AI) and ML systems have advanced significantly in recent years AI is not intelligent in the biological sense. These systems are limited to performing only those tasks for which they have been specifically programmed and trained, and are inherently subject to safety hazards when encountering situations outside them. The issue is further limiting to DoD applications, where situations can be unpredictable and the ability to react quickly and adapt to dynamic circumstances is of primary importance.
The Lifelong Learning Machines (L2M) program at DARPA seeks to achieve paradigm-changing developments in AI architectures and ML techniques. The program seeks to develop systems that can learn continuously during execution and become increasingly expert while performing tasks, are subject to safety limits, and apply previous skills and knowledge to new situations - without forgetting previous learning.
L2M consists of two technical areas. The first concentrates on the development of complete systems and their components; the second brings together researchers with diverse expertise to explore biological mechanisms that underlie learning, which will be translated into a new generation of computational architectures, mechanisms, and algorithms. Discoveries in both technical areas are expected to generate new methodologies that will allow AI systems to learn and improve during tasks, apply previous skills and knowledge to new situations, incorporate innate system limits, and enhance safety in automated assignments.
Contemporary deep learning has enabled a next generation of artificial intelligence applications, opening the door to potential breakthroughs in many aspects of our lives. Recently, Pacific Northwest National Laborator (PNNL) researchers teamed up to explore how deep learning could help interpret signals from radioactive decay events. Analyzing data collected in PNNL’s Shallow Underground Laboratory, tests showed that the algorithm separated signal events from instrument “noise” with nearly 100 percent accuracy. Findings from this research could indicate underground nuclear testing, for example, or a rapidly depleting aquifer.
Data collected from systems—ranging from field sensors to commercial vendor data services—often are incomplete and require specialized software to be usable. PNNL has extensive experience augmenting metadata and applying semantic technologies and other processing techniques to provide additional context and meaning for data. PNNL experience spans local data access to highly distributed data collections, as well as traditional storage systems to high-performance warehousing technologies and cloud-based solutions.
In the energy sector, systems are changing rapidly as technology evolves and new sources of energy generation come to fruition. VOLTTRONTM is a platform for integrating software and smart device operations and communications over the power grid. It provides an environment where a wide range of data and devices connect seamlessly and securely to make decisions based on user needs and preferences – improving system performance in buildings, and creating a more flexible and reliable grid.
Data analytic technologies are also having an important impact in the biomedical field. The discovery of cancer biomarkers offers the prospect of earlier diagnosis, more accurate prognosis, effective monitoring of therapeutic response, and the opportunity to target patient treatment. P-Mart (Panomics Marketplace: analyses via reproducible techniques) is an interactive web-based tool for biomedical scientists to analyze global proteomics data, enabling the analysis and integration of omics data in a robust and reproducible manner.
PNNL data analytics and machine learning methods – including capabilities in graph, streaming, and high-performance data analytics – are separating useful data from noise, increasing predictive power and efficiency of computational models, and aiding in decision making to address critical national and global issues. PNNL researchers apply scientific and mathematical techniques to develop novel, scalable algorithms designed to run on leadership-class facilities or commodity hardware.
As the amount of data being produced, manipulated, and stored exponentially increases, so does the very real threat of cyber-security breaches and fraud. Meanwhile, federal budgets and staff resources continue to decrease. ML can provide high-value services for federal agencies including data management and analytics, security threat detection, and process improvement—but the list does not stop there. ML is a type of AI that takes human-input data, analyzes it, and learns from it. Three types of learning can occur: supervised learning in which the machine analyzes past high quality data and makes decisions about future data with the learned knowledge, unsupervised learning in which the machine makes inferences about future data based on patterns it finds within past data, and a combination of the two.
According to a recent MeriTalk survey, 81% of feds are currently utilizing some form of Big Data analytics for cybersecurity, while only 45% found their efforts to be highly effective. These numbers are staggering considering that Big Data is still a relatively new discipline to most people. Google Trends Analyses show that traditional Big Data is being phased out just as fast as it initially exploded, and that it will soon be replaced with AI applications and ML.
But not to fear, ML will not replace humans; not yet anyway. This is where data scientists come in. Data scientists are a critical component of ML for analytics and data-based predictions. Data scientists conduct statistical and algorithm modeling, and determine which ML platform is best suited for the data. R and Python are currently the two most popular programming languages in ML. More importantly, the data scientist must determine what the machine will do with the data. Innovative companies such as Yelp, Facebook, and Google have already fully implemented ML into their platforms. For example, for a company like Yelp, the site’s tens of millions of images ARE its data set. Highly skilled data scientists must first teach the bot how to classify the existing images as well as rules for classifying future images.
Of course, most federal agencies wouldn’t consider images a major component of their critical data; however the same concepts can be applied to whatever data needs to be classified, analyzed, secured, or visualized.
As AI and ML technologies improve, the need for highly trained data scientists will only increase. Very soon, machines will have the ability to conduct more accurate analysis with even less data, but this will only be possible with expert statistical modeling and perfected algorithms created by data scientists. According to Dr. Heather Benz, Applied Biomedical Engineer, Johns Hopkins; “We are teaching more and more of our engineering students about how to design next-generation technologies that incorporate ML. There will only be an increase in need for individuals who can understand, design, and leverage these tools. They have broad applications in everything from business intelligence to consumer electronics to medical devices, but there is a lot of nuance to how they’re built, used, and validated.”