Research map / Reinforcement learning / Visual and representation RL
Understanding Plasticity in Neural Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add layer normalization after each hidden layer | Double DQN on Arcade Learning Environment · 57 Atari games | performance (human-normalized) | Helped |
| add layer normalization after each conv and fully-connected layer | CNN and MLP on classification MDP probe tasks · classification MDPs (CIFAR-10/MNIST) | plasticity loss | Helped |
| categorical two-hot output representation instead of regression | Transformer, CNN, MLP on classification MDPs · classification MDPs (CIFAR-10) | plasticity loss | Helped |
| reset last layer at each target network update | Classification MDP agents · classification MDPs | plasticity loss | Helped |
| shrink and perturb at each target update | Classification MDP agents · classification MDPs | plasticity loss | No clear effect |
| weight decay 1e-5 | Classification MDP agents · classification MDPs | plasticity loss | No clear effect |
| increase Adam epsilon and lower second-moment decay (b2=0.9) | Adam on non-stationary MNIST memorization · MNIST random-label re-randomization | accuracy / divergence | Helped |
| adding layer normalization to CNN | CNN regression on classification MDPs · classification MDPs | Hessian spectral norm sharpness | Helped |
| weight norm, feature rank, weight rank, dead units as predictors of plasticity loss | DQN agents across tasks and observation spaces · classification MDPs | correlation with plasticity | Mixed |
Closest papers by meaning
- A Study of Plasticity Loss in On-Policy Deep Reinforcement Learning (NeurIPS 2024 · similarity 0.92)
- Revisiting Plasticity in Visual Reinforcement Learning: Data, Modules and Training Stages (ICLR 2024 · similarity 0.90)
- Mitigating Plasticity Loss in Continual Reinforcement Learning by Reducing Churn (ICML 2025 · similarity 0.89)
- Neuroplastic Expansion in Deep Reinforcement Learning (ICLR 2025 · similarity 0.87)
- Learning the Plasticity: Plasticity-Driven Learning Framework in Spiking Neural Networks (NeurIPS 2025 · similarity 0.87)
- Rethinking the Stability-Plasticity Trade-off in Continual Learning from an Architectural Perspective (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.