When Does Backpropagating Through Policy Memory Matter? Physical Credit, Optimizer Updates, and Observability
cs.LG, cs.RO
Submitted: 2026-09-27
Updated: 2026-10-03
Terminology
Sources
- Delayed Optimizer-State Transport Shapes Short-Horizon Training Decisions
- Memo: Training Memory-Efficient Embodied Agents with Reinforcement Learning
- Dream to Control: Learning Behaviors by Latent Imagination
- Learning Quadrotor Control From Visual Features Using Differentiable Simulation
- Memory-based control with recurrent neural networks
- Sparse Attentive Backtracking: Temporal CreditAssignment Through Reminding
- Gradients are Not All You Need
- Autonomous Vehicle Controllers From End-to-End Differentiable Simulation
- When Do Transformers Shine in RL? Decoupling Memory from Credit Assignment
- Unbiasing Truncated Backpropagation Through Time
- Accelerated Policy Learning with Parallel Differentiable Simulation
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks