In computer science, machine learning has great potential to transform the way we make predictions and observations for real-world applications, but this depends on our ability to interpret and use raw data practically, seamlessly integrating it into our workflow processes.
Dr. Una-May O’Reilly’s research goal is to improve the application of machine learning, so engineers and scientists of complex systems can gain improved insights toward serving others. With her research group, AnyScale Learning for All (ALFA), she develops new data-driven analyses of online coding courses, deep learning techniques for program representations, adversarial attacks on machine learning models, model training for adversarial robustness, and cyber hunting tools and cyber arms race models.
With this data-driven approach, Dr. O’Reilly, who holds a PhD from Carleton University in Ottawa, Canada and joined CSAIL in 1996, and her group are exploring ways to exploit state-of-the-art artificial intelligence and scalable machine learning for applications in various enterprises, including healthcare, online learning, and security.
She was first drawn to this area of research through her highly regarded and award-winning work on evolutionary algorithms, which she found were a good fit for the evolving arms race between tax avoidance and tax loophole repair. This research led to her group looking at vulnerabilities in a subsection of the Internal Revenue code and devising algorithms to automatically exploit loopholes and avoid auditing, even under co-adaption.
Now one of the main areas Dr. O’Reilly is investigating is cybersecurity and how to stop destructive and escalating arms races. Industry and governments are increasingly relying on AI and machine learning systems to handle sophisticated tasks, but the security of the AI systems is at risk. The systems have inherent vulnerabilities, thus adversaries can attack them. To understand these attacks and plan our defenses, we have to think like attackers and anticipate attacks when we train models. Dr. O’Reilly hopes to understand the nature of adversarial intelligence by computationally replicating it—that is, by developing Artificial Adversarial Intelligence. Artificial Adversarial Intelligence will help reveal the dynamics of conflicting behavior and how adaptation drives it, so that we can put a stop to these types of arms races.
Another area she is exploring is new data-driven analyses of online coding courses. Dr. O’Reilly, who is interested in the nature of understanding computer programs in general, is targeting the activity of learners when they learn online. In order to understand what is working or not working for online learners, she and her ALFA team are able to collect clicks and trace the navigation of learners through course material. They use data science and machine learning to analyze learner behavior, and these insights can be fed back to learning designers to improve teaching.
Her research also dives into deep learning techniques for program representations. Typically, programming code varies in length, obeys syntax, and expresses semantics. To automatically detect bugs or code malware, they have to develop a computer-friendly representation of code that encompasses all of these properties. This representation can be used by software enterprises in machine learning detectors and classification training.
Dr. O’Reilly’s data-driven, application-focused approach to AI and machine learning will lead to more secure systems, more effective learning online, and optimized automatic programming and program comprehension across industries and organizations.
Una-May is leader of the AnyScale Learning For All (ALFA) group at CSAIL. ALFA focuses on scalable machine learning, evolutionary algorithms, and frameworks for large scale knowledge mining, prediction and analytics. The group has projects in cybersecurity, healthcare, and online education.
Una-May was awarded the EvoStar Award for Outstanding Contribution of Evolutionary Computation in Europe in April, 2013. She is also is a Fellow of the International Society of Genetic and Evolutionary Computation, now ACM Sig-EVO.
Una-May co-founded ACM SigEVO in 2004. She serves as Vice-Chair of ACM SigEVO. In 2013 she inaugurated the Women in Evolutionary Computation group at GECCO.
Una-May served as chair of the largest international Evolutionary Computation Conference, GECCO, in 2005. She has served on the GECCO business committee, co-led the 2006 and 2009 Genetic Programming: Theory to Practice Workshops and co-chaired EuroGP, the largest conference devoted to Genetic Programming.
Una-May serves as the area editor for Data Analytics and Knowledge Discovery for Genetic Programming and Evolvable Machines (Kluwer), as editor for Evolutionary Computation (MIT Press), and as action editor for the Journal of Machine Learning Research.
Una-May has a patent for a original genetic algorithm technique applicable to internet-based name suggestions.
Una-May holds a B.Sc. from the University of Calgary, and a M.C.S. and Ph.D. (1995) from Carleton University, Ottawa, Canada. She joined the Artificial Intelligence Laboratory, MIT as a Post-Doctoral Associate in 1996.
The MOOC Learner Project provides learning scientists, instructional designers and online education specialists with open source software that enables them to efficiently extract teaching and learning insights from the data collected when students learn on the edX or open edX platform.
Our goal is to understand the nature of cyber security arms races between malicious and bonafide parties. Our vision is autonomous cyber defenses that anticipate and take measures against counter attacks.
This CoR brings together researchers at CSAIL working across a broad swath of application domains. Within these lie novel and challenging machine learning problems serving science, social science and computer science.
This CoR aims to develop AI technology that synthesizes symbolic reasoning, probabilistic reasoning for dealing with uncertainty in the world, and statistical methods for extracting and exploiting regularities in the world, into an integrated picture of intelligence that is informed by computational insights and by cognitive science.
When a power company wants to build a new wind farm, it generally hires a consultant to make wind speed measurements at the proposed site for eight to 12 months. Those measurements are correlated with historical data and used to assess the site’s power-generation capacity.This month CSAIL researchers will present a new statistical technique that yields better wind-speed predictions than existing techniques do — even when it uses only three months’ worth of data. That could save power companies time and money, particularly in the evaluation of sites for offshore wind farms, where maintaining measurement stations is particularly costly.
By crunching 130 million mouse-clicks, two CSAIL researchers have developed a machine-learning model that can predict with surprising accuracy whether or not a MOOC student will drop out of a given course.