Michael Harradon,
Senior Scientist

Mr. Harradon holds a M.Eng. in Computer Science and a B.S. in Physics and Computer Science, both from MIT, where he also won first place in the 2016 100K Pitch Competition. His primary research interests are machine learning and reinforcement learning (RL), and he is an expert in probabilistic AI, including probabilistic programming. He’s applied novel machine learning approaches to a range of domains, including financial data analysis, radar signals processing, and computer vision. Other areas of expertise include signals analysis, game theory, optimization methods, and genetic algorithms.

Harradon is the inventor of Neural Program Policies (NPPs) and a pioneer in the NPP approach to explainable-by-construction RL agents. As principal investigator for several projects funded under DARPA’s OFFSET program, he is applying NPPs to automatically learn drone swarm behaviors in urban settings. For PERCIVAL-MERLIN, he used NPPs to develop RL-trained agents that predict and track surface vessels.

Harradon has led numerous other projects at Charles River Analytics. As the principal investigator on the STAPLES project (under the DARPA LogX program), he is leading the development of new probabilistic models and inference algorithms for logistics forecasting.

Harradon is a key figure in Charles River’s leadership in the emerging field of explainable AI; he works on explainable modeling and deep RL for the CAMEL effort and he developed novel approaches to identifying human intelligible concepts for the ALPACA project. He is the co-author (with Jeff Druce and Brian Ruttenberg) of “Causal Learning and Explanation of Deep Neural Networks Via Autoencoded Activations” and co-author (with Jeff Druce and James Tittle) of  “Explainable Artificial Intelligence (XAI) for Increasing User Trust in Deep Reinforcement Learning Driven Autonomous Systems.”