Trajectory Generalization

Trajectory generalization in machine learning, particularly within reinforcement learning and deep neural networks, focuses on improving the ability of models to apply learned knowledge to unseen situations or data distributions. Current research emphasizes leveraging offline data, including pre-training with imitation learning and augmenting training data with simulated trajectories generated by models like world transformers, to enhance generalization. This area is crucial for improving the sample efficiency and robustness of AI systems, enabling their deployment in real-world scenarios where complete exploration is impractical or impossible.

Papers

November 7, 2024

DINO-WM: World Models on Pre-trained Visual Features enable Zero-shot Planning
Gaoyue Zhou, Hengkai Pan, Yann LeCun, Lerrel Pinto
Zero Shot Pre Trained Task Planning World Model Task Agnostic Trajectory Generalization

October 21, 2024

FlickerFusion: Intra-trajectory Domain Generalizing Multi-Agent RL
Woosung Koh, Wonbeen Oh, Siyeol Kim, Suhin Shin, Hyeongjin Kim, Jaein Jang, Junghyun Lee, Se-Young Yun
Multi Agent Reinforcement Learning Domain Generalization Multi Agent Particle Multi Agent RL Inference Stage Trajectory Generalization

April 16, 2024

Offline Trajectory Generalization for Offline Reinforcement Learning
Ziqi Zhao, Zhaochun Ren, Liu Yang, Fajie Yuan, Pengjie Ren, Zhumin Chen, jun Ma, Xin Xin
Offline Reinforcement Learning Trajectory Wise Reward Trajectory Generalization

June 4, 2023

Bad Habits: Policy Confounding and Out-of-Trajectory Generalization in RL
Miguel Suau, Matthijs T. J. Spaan, Frans A. Oliehoek
Reinforcement Learning Agent Handwritten Trajectory Human Like RL Confounding Bias Habitual Behavior State Action Trajectory Generalization

April 25, 2023

Learning Trajectories are Generalization Indicators
Jingwen Fu, Zhizheng Zhang, Dacheng Yin, Yan Lu, Nanning Zheng
Strong Generalization Generalization Error DNN Training Generalization Measure Trajectory Generalization

April 18, 2023

Using Offline Data to Speed-up Reinforcement Learning in Procedurally Generated Environments
Alain Andres, Lukas Schäfer, Esther Villar-Rodriguez, Stefano V. Albrecht, Javier Del Ser
Reinforcement Learning Imitation Learning Optimal Policy Environment Feature Offline Data Procedural Text Trajectory Generalization

June 9, 2022

Trajectory-dependent Generalization Bounds for Deep Neural Networks via Fractional Brownian Motion
Chengli Tan, Jiangshe Zhang, Junmin Liu
Deep Neural Network Generalization Gap Gradient Noise Rademacher Complexity Stochastic Recursive Gradient Fractional Brownian Motion Norm Based Generalization Bound Trajectory Generalization