John Winder Email & Phone Number

I lead the Advanced Artificial Intelligence Algorithms section at the Johns Hopkins University Applied Physics Laboratory (JHU/APL), where we research deep reinforcement learning (RL). In my role as a Senior Staff Research Scientist, I develop novel machine learning (ML) algorithms for real-world systems such as aircraft, satellites, and complex software agents.At JHU/APL, I supervise a team researching multi-agent RL in cooperative-competitive settings. I'm developing new methods combining RL with attention (transformers) and relational reasoning (graph neural networks, GNNs). Creating decision-making agents, agents that reason about long-term goals under uncertainty, genuinely excites me and serves as the objective guiding my team's research.My journal and conference publications have appeared in TAAS, NeurIPS, AAAI, IJCAI, ICAPS, and similar venues. Research I've conducted spans a breadth of AI/ML subtopics from deep RL to grounded language acquisition (computer vision + natural language processing / NLP). My experience and current interests include scalable decentralized multi-agent RL systems, large language models (LLMs, foundation models), reinforcement learning from human feedback (RLHF), self-attention and neural memory mechanisms, diffusion models and generative adversarial networks (GANs), geometric deep learning, deep learning theory, semantic segmentation, contextual bandits, and interpretable / explainable ML and RL.In 2019, I received my Ph.D. in Computer Science from UMBC, where I specialized in hierarchical reinforcement learning (HRL), exploring temporal and state abstractions for model-based RL and probabilistic planning.Since joining JHU/APL, I have proposed projects competitively awarded over $2MM in internal research and development funding, focused on high-risk, high-impact AI/ML technologies. I also make key contributions to innovative AI/ML programs on behalf of organizations such as DARPA, AFRL, and NAVAIR. For example, my RL research has supported the DARPA Air Combat Evolution (ACE) program, including the AlphaDogfight Trials event where a dogfighting AI agent defeated a human fighter pilot 5-0: the livestreamed final event has received over 400K views on YouTube, and the story has received international coverage including in the New Yorker, Popular Science, and the 2020 State of AI Report.