Sparse-TD-JEPA: Improving Zero-Shot Reinforcement Learning With Sparse Latent Task Representations

Name: Sparse-TD-JEPA: Improving Zero-Shot Reinforcement Learning With Sparse Latent Task Representations
Uploaded: 2026-10-18T12:06:03.480313+00:00
Duration: 226 s
Description: Sparse-TD-JEPA introduces an L1 sparsity penalty during training for a task encoder's latent representations. This forces the reward-predictive structure into a smaller subspace, concentrating reward signals. This results in significantly improved zero-shot reinforcement learning performance across various benchmarks by creating sharper, more goal-focused reward landscapes.

Sparse-TD-JEPA introduces an L1 sparsity penalty during training for a task encoder's latent representations. This forces the reward-predictive structure into a smaller subspace, concentrating reward signals. This results in significantly improved zero-shot reinforcement learning performance across various benchmarks by creating sharper, more goal-focused reward landscapes.

Open the full episode