Adversarial Cyber Security


A cyber security ecosystem is characterized by intelligent, adaptive adversaries. Defenders engage in an arms race with attackers as both sides take turns crafting new responses to each other’s actions. Adversarial arms races occur in multiple cyber domains. Domains include networks, malware detection, regulatory circumvention, data exfiltration, capture-the-flag exercises and even the decades-running "War in Memory" in C/C++. Problematically, the current paradigm cannot handle the scale, severity and adaptive strategy of forthcoming threats. Defenses are largely reactive. Each new attack typically requires identification, human response, and design intervention to prevent it. Our research questions revolve around how to develop autonomous, proactive cyber defenses that are anticipatory and adaptable to counter attacks. We have a variety of ongoing projects: The Stealth Cyber Security Project, The CASCADE Enclave Modeling Project, and Malware Development and Detection Arms Races.

MOOC Learner Project: Advancing learning behavior analytics through Data Science


Each time a learner interacts with an e-learning system it is possible to capture a record of their engagement. Data comprising mouse clicks, video controls, problem responses, programming, collaborations and discussions then becomes available to learning science. MOOC Learner Project’s goal is to tap into the immense potential of this data to provide insights into how students learn and how instructors can effectively teach. The challenge is to provide technology and develop new approaches that transforms this fundamentally different set of observations into actionable knowledge.

Gigabeats: Data science for medical sensor data


Our projects tap into machine learning to interpret and exploit repositories holding waveform data i.e. arterial blood pressure, ECG, and EEG. They include BeatDB where we are developing a fast and scalable framework for compiling machine learning data sets from waveform repositories and Trajectories Like Mine where we are developing a sub-linear time method called Locality Sensitive Hashing to find approximate nearest neighbors in the waveform space. Our goal is to help medical researchers and clinicians understand the growing repositories of waveform and signal data collected from critically ill patients. These projects are generously supported by Philips Research North America

STEALTH Tax Project 


Cyberspace has become a competition ground occupied by intelligent, adaptive adversaries. Defenders and attackers engage in arms races as both sides take turns crafting new responses to each other’s actions. The arms races play out in multiple cyber arenas ranging from networks sustaining Denial of Service attacks, to compromised enterprise systems being profiled by internal reconnaissance, to anti-virus detectors encountering unanticipated malware. Current defenses are largely reactive — each new attack typically requires identification, human response, and design intervention to prevent it. They are inadequate to address the ever increasing scale, severity and adaptive strategies of malicious parties. Our vision is autonomous cyber defenses that anticipate and take measures against counter attacks. Our technical approach frames a robust optimization problem where the objectives of the two sides conflict and the positive gains of one side imply negative outcomes for the other. The co-optimization problem is solved with co-evolutionary algorithms. We have several ongoing projects: Rivals: More robust and resilient networks under extreme DDoS attacks
, ADHD: Adversarial Dynamics when Harnessing Deception, The Arms Race of Anti-Virus Detection and Viruses, STEALTH: Understanding the Relationship between Tax Non-Compliance and Tax Law

FlexGP: Flexible ML with Genetic Programming 


Genetic programming is a mature, robust multi-point search technique (inspired by evolution) which supports readable, and flexibly specified learning representations which can readily express linear or non-linear data relationships. It is well suited to parallelization and machine learning. It has a strong record in real world domains. In a nutshell, the FlexGP project goal is scalable machine learning using genetic programming (GP).


Past Projects


Highlights of low activity and past projects: Scalable EDA-GP, The Human Data Interaction Project, Wind Energy, AutotuningHierarchical Genetic Algorithms for Parallelization of Sparse Matrix Algebra, Networks, Meta-Optimization: Improving Compilation with Genetic Programming, Support Vector Machines: Performance Analysis, Multi-Objective Optimization Algorithm Design, Hybrid Machine-Learning and Optimization, Analog Reconfigurable Systems, Adaptive Resource Allocation, Evolvable Hardware