technique

Elastic Weight Consolidation

techniqueactiveelastic-weight-consolidation-bd20d3c4·1 events·first seen 1mo ago

Aliases: Elastic Weight Consolidation

Co-occurring entities

Dark Experience Replay University of California Los Angeles LIBERO catastrophic forgetting GRPO LoRA Sony Jiaheng Hu Nanyang Technological University Jay Shim OpenVLA-OFT University of Texas Austin

More like this (12)

Extrapolative Weight Averaging Lean EVA Ecom-RLVE Delta Weight Sync ESI-Bench Nemotron Elastic Global-batch Load Balancing Reversible Residual Layers Conservation X Labs ZEDA (Zero-Expert Self-Distillation Adaptation)Model Collapse

Recent events (1)

5The Batch·1mo ago·source ↗

Sony and University Researchers Train Robots To Learn Without Catastrophic Forgetting

Researchers from UT Austin, UCLA, Nanyang Technological University, and Sony developed a sequential fine-tuning recipe combining LoRA and on-policy reinforcement learning (GRPO) to reduce catastrophic forgetting in vision-language-action (VLA) models for robotics. Applied to the OpenVLA-OFT model on the LIBERO benchmark, the method achieved 81.2% success on libero-spatial tasks with near-zero forgetting (0.3 percentage point drop), outperforming established continual learning baselines including Dark Experience Replay and Elastic Weight Consolidation. The approach requires no replay of prior task data and also showed modest generalization to unseen tasks. The authors note the method has not yet been tested outside robotics simulation contexts.

Evaluation and Benchmarking Agent and Tool Ecosystem Elastic Weight Consolidation Dark Experience Replay University of California Los Angeles +11 more