NeurIPS 2022 brought Furong Lab to New Orleans with work across a deliberately broad question: how can learning systems remain useful when their environment, data, users, or collaborators do not behave exactly as expected?
The program included six papers in the main conference and nine workshop papers. “Controllable Attack and Improved Adversarial Training in Multi-Agent Reinforcement Learning” also received an outstanding paper award at the Trustworthy and Socially Responsible Machine Learning workshop.
Fairness that survives change
Several projects studied fairness beyond a single fixed dataset. The work asked how consistency, invariance, and long-term decision effects can help equitable behavior survive distribution shifts and sequential feedback.
Robust sequential decision-making
Another cluster examined reinforcement-learning agents facing adaptive adversaries, perturbed observations, and strategic multi-agent interactions. These papers treated robustness as an ongoing decision problem rather than a one-time training constraint.
Efficient and reliable learning
The remaining work connected distributed training, model invariance, representation learning, and optimization. Together, the papers reflected a lab-wide theme that still shapes our current program: capability and trustworthiness must be designed together.