Research map / Continual learning / Online and replay-based

Prevalence of Negative Transfer in Continual Reinforcement Learning: Analyses and a Simple Baseline

ICLR 2025 · Continual learning › Online and replay-based · approach: Plasticity-preserving continual reinforcement learning

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Shows that negative transfer to newly arriving tasks is prevalent in continual reinforcement learning and that existing plasticity-loss methods do not fix it, and proposes a simple baseline.

Problem

Negative transfer when a new task arrives in continual reinforcement learning (CRL), which current CRL methods, including those targeting plasticity loss, do not effectively address.

Approach

Argues negative transfer on new-task arrival is a neglected problem in CRL, provides comprehensive experimental analysis showing it is frequent, and introduces a simple baseline method to handle it.

Surprising findings

• Negative transfer frequently occurs in CRL and is not effectively addressed by recent methods for mitigating plasticity loss.

Experimental setup

Baselines: methods for mitigating plasticity loss in CRL

Closest papers by meaning

• Continual Reinforcement Learning by Planning with Online World Models (ICML 2025 · similarity 0.89)
• Continual Knowledge Adaptation for Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
• Ergodic Risk Measures: Towards a Risk-Aware Foundation for Continual Reinforcement Learning (NeurIPS 2026 · similarity 0.87)
• Mitigating Plasticity Loss in Continual Reinforcement Learning by Reducing Churn (ICML 2025 · similarity 0.87)
• Blending Imitation and Reinforcement Learning for Robust Policy Improvement (ICLR 2024 · similarity 0.87)
• Dual RL: Unification and New Methods for Reinforcement and Imitation Learning (ICLR 2024 · similarity 0.87)