Philip Thomas

Philip Thomas is an Associate Professor and Doctoral Program Director at the Manning College of Information and Computer Sciences, University of Massachusetts Amherst. He leads the Autonomous Learning Lab (ALL) and co-founded the Reinforcement Learning Conference (RLC). His research focuses on reinforcement learning, AI safety, and algorithms that ensure safety guarantees for high-risk applications like healthcare and digital marketing.

Education:

• PhD in Computer Science, University of Massachusetts Amherst (2015)
• MSc in Computer Science, Case Western Reserve University (2009)
• BSc in Computer Science, Case Western Reserve University (2008)

Research Interests: Thomas specializes in designing biologically plausible reinforcement learning algorithms and ensuring safety through frameworks like Qualia Optimization and Seldonian Algorithms. His work emphasizes off-policy evaluation, fairness guarantees, and ethical AI. Recent projects include developing benchmarks for medical decision-making (e.g., ICU-Sepsis) and analyzing adversarial robustness in speech denoising models.

Articles Trends: His recent work spans high-confidence policy evaluation, fairness metrics, and algorithmic safety. Key themes include improving benchmarking practices, rethinking eligibility traces, and leveraging state abstraction for consistent off-policy evaluation.

Awards & Grants:

• Armstrong Award
• Co-PI on Army Research Grant (IoBT), NSF grant (FMitF)
• Significant funding from Adobe Research

Advising & Grants: Thomas has overseen grants totaling millions and mentored students in reinforcement learning and AI safety. His current focus includes exploring qualia optimization for doctoral applications (2026-2027).

Labs & Teams: He directs the Autonomous Learning Lab and collaborates on interdisciplinary projects at the Center for Data Science, emphasizing ethical AI and safe machine learning systems.