t-DGR: A Trajectory-Based Deep Generative Replay Method for Continual Learning in Decision Making
This addresses the problem of compounding errors in trajectory generation for continual learning, offering a scalable solution for AI agents that need to learn multiple tasks sequentially, though it is incremental as it builds on existing deep generative replay methods.
The paper tackles catastrophic forgetting in continual learning for decision-making tasks by proposing a non-autoregressive deep generative replay method that generates trajectories conditioned on timestep, achieving state-of-the-art performance on Continual World benchmarks with improved average success rates.
Deep generative replay has emerged as a promising approach for continual learning in decision-making tasks. This approach addresses the problem of catastrophic forgetting by leveraging the generation of trajectories from previously encountered tasks to augment the current dataset. However, existing deep generative replay methods for continual learning rely on autoregressive models, which suffer from compounding errors in the generated trajectories. In this paper, we propose a simple, scalable, and non-autoregressive method for continual learning in decision-making tasks using a generative model that generates task samples conditioned on the trajectory timestep. We evaluate our method on Continual World benchmarks and find that our approach achieves state-of-the-art performance on the average success rate metric among continual learning methods. Code is available at https://github.com/WilliamYue37/t-DGR.