Alfredo Garcia’s Research
Overview
My research asks how systems of interacting decision-makers — people, organizations, and increasingly AI agents — learn, respond to incentives, and can be designed to work well together. I began by studying strategic behavior in engineered markets and networks, using game theory, mechanism design, and distributed optimization to coordinate power grids and wireless networks. Today my group brings the same questions to AI: recovering what people want from how they act — learning reward models from demonstrations and feedback to align large language models — and modeling human attention and decision-making so that people and automation can work together safely.
-
AI Alignment and Learning from Human Feedback
Learning what people want from how they act, and using it to align language models and reinforcement learning agents.
-
Human-AI Teaming and Cognitive Modeling
Bayesian and active-inference models of human attention, effort, and control, applied to driving and automation.
-
Distributed Optimization and Control
Algorithms with provable guarantees for networks of agents that learn and optimize without a central coordinator.
-
Markets, Mechanisms, and Engineering Systems
Game theory, mechanism design, and multi-agent learning for power grids, communication networks, and operations.
Research Areas
How can an AI system learn what people want from how they behave? We develop inverse reinforcement learning methods with finite-sample guarantees — a maximum-likelihood framework for offline inverse RL, robust Bayesian inverse RL under model misspecification, and analysis of the overparameterized regime — and carry them into large language model alignment. Our results show that reward models learned from human demonstrations improve supervised fine-tuning, and that learning rewards and policies jointly from demonstrations and preference feedback outperforms either alone. Current work extends these ideas to long-horizon LLM agents and robust reinforcement learning.
Selected publications
- Ch. Li, S. Zeng, Z. Liao, J. Li, D. Kang, A. Garcia, and M. Hong. “Joint Reward and Policy Learning with Demonstrations and Human Feedback Improves Alignment.” ICLR, 2025.
- J. Li, S. Zeng, H. Wai, Ch. Li, A. Garcia, and M. Hong. “Getting More Juice Out of the SFT Data: Reward Learning from Human Demonstration Improves SFT for LLM Alignment.” NeurIPS, 2024.
- S. Zeng et al. “Aligning Large Language Models with Human Feedback: Mathematical Foundations and Algorithm Design.” IEEE Signal Processing Magazine, 2026.
- S. Zeng, Ch. Li, A. Garcia, and M. Hong. “Maximum Likelihood Inverse Reinforcement Learning with Finite-Time Guarantees.” NeurIPS, 2022.
Safe human-AI teaming depends on knowing what the person in the loop is attending to and how they will respond. We build computational models of attention, effort, and fatigue — using Bayesian inference and active inference — and estimate them from behavioral data. Applications include driver distraction and secondary-task engagement, car-following, and takeovers from automated vehicles, with implications for driver-assistance design and handoff protocols.
Selected publications
- L. Kashyap, Z. Wang, Y. Chang, M. Zahabi, and A. Garcia. “Inferring Hidden Attentional States in Driving: A Bayesian Approach to Modeling Distraction and Secondary Task Engagement.” Human Factors, 2026.
- L. Kashyap, Z. Wang, Y. Chang, and A. Garcia. “A Structural Model of Attentional Effort Dynamics: Evidence from a Naturalistic Discrimination Task.” Human Factors, 2025.
- R. Wei, A. Garcia, A. McDonald, G. Markkula, J. Engström, and M. O’Kelly. “Learning an Active Inference Model of Driver Perception and Control: Application to Vehicle Car-Following.” IEEE Transactions on Intelligent Transportation Systems, 2025.
- Z. Wang, Y. Chang, B. J. Schmeichel, and A. Garcia. “The Effects of Mental Fatigue on Effort Allocation: Modeling and Estimation.” Psychological Review, 129(6):1457–1485, 2022.
Many learning and control problems must be solved by networks of agents that cannot pool their data. We design decentralized algorithms with provable convergence, including methods for non-convex and manifold-constrained problems — such as optimization over the Stiefel manifold, which underlies PCA, subspace tracking, and orthogonality constraints in neural networks — and networked learning with correlated, privately held data. Earlier work showed that swarming, a bio-inspired form of flocking among agents, provably accelerates stochastic optimization by reducing noise.
Selected publications
- Y. Sun, S. Chen, A. Garcia, and S. Shahrampour. “Retraction-Free Decentralized Non-Convex Optimization with Orthogonal Constraints.” IEEE Transactions on Automatic Control, 2026.
- S. Chen, A. Garcia, M. Hong, and S. Shahrampour. “On the Local Linear Rate of Consensus on the Stiefel Manifold.” IEEE Transactions on Automatic Control, 2023.
- S. Chen, A. Garcia, M. Hong, and S. Shahrampour. “Decentralized Riemannian Gradient Descent on the Stiefel Manifold.” ICML, 2021.
- S. Pu and A. Garcia. “Swarming for Faster Convergence in Stochastic Optimization.” SIAM Journal on Control and Optimization, 56(4):2997–3020, 2018.
When independent participants share infrastructure, their strategic behavior shapes the outcome. We use game theory, mechanism design, and stochastic dynamic programming to allocate capacity efficiently under congestion, price interference in wireless networks, and plan capacity under uncertain demand. Recent work applies multi-agent reinforcement learning and distributed optimization to power systems — multi-area power exchange and unit commitment with energy storage — enabling coordination without a central operator.
Selected publications
- S. Biswas, B. Cavdar, A. Garcia, and J. Geunes. “The Price of Flexibility in Electricity Markets.” Energy Economics, 2025.
- A. Garcia, R. Khatami, C. Eksin, and F. Sezer. “An Incentive Compatible Iterative Mechanism for Coupling Electricity Markets.” IEEE Transactions on Power Systems, 37(2):1241–1252, 2022.
- J. Barrera and A. Garcia. “Auction Design for the Efficient Allocation of Service Capacity Under Congestion.” Operations Research, 63(1):151–165, 2015.
- A. Garcia and J. Shen. “Equilibrium Capacity Expansion under Stochastic Demand Growth.” Operations Research, 58(1):30–42, 2010.