Research map / Architectures and efficiency / RNNs and normalization
Hebbian and Gradient-based Plasticity Enables Robust Memory and Rapid Learning in RNNs
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 plasticity to an RNN | RNN · CIFAR-FS and miniImageNet 5-way 1-shot with Conv-4 | test accuracy: 39.9 $\pm$ 0.8 -> 55.5 $\pm$ 1.0; 44.1 $\pm$ 0.7 -> 49.8 $\pm$ 0.5 | Helped |
| Use Hebbian plasticity instead of no plasticity | RNN · CIFAR-FS and miniImageNet 5-way 1-shot with ResNet-12 | test accuracy: 41.5 $\pm$ 2.4 -> 59.6 $\pm$ 1.5; 42.3 $\pm$ 0.6 -> 50.4 $\pm$ 1.1 | Helped |
| Use gradient-based rather than Hebbian plasticity | Plastic RNNs and LSTMs · Few-shot regression, Linear and MLP mappings | test error: .378 $\pm$ .059 vs .301 $\pm$ .001; .200 $\pm$ .042 vs .108 $\pm$ .001; .165 $\pm$ .001 vs .142 $\pm$ .001; .077 $\pm$ . | Helped |
| Use Hebbian plasticity instead of no plasticity | LSTM · Few-shot regression, MLP mapping, K = 10 | test error: .241 $\pm$ .043 -> .282 $\pm$ .013 | Hurt |
| Use an adaptive internal learning rate rather than a fixed one (ablation) | Hebbian and gradient-based plastic RNNs · Copying task | MSE loss | Helped |
| Set max_norm to 100 instead of 1 (ablation) | Gradient-based plasticity · Copying task, n = 5 and m = 40 | training stability: 3 out of 4 | Hurt |
| Use connection-specific learning rates instead of a fixed global learning rate (ablation) | Hebbian and gradient-based plastic RNNs · Cue-reward association | validation loss | Helped |
| Add the auxiliary output vector ỹ_t (ablation) | Gradient-based plasticity · Cue-reward association | validation error | Helped |
| Increase dim(ỹ_t) beyond 4 (ablation) | Gradient-based plasticity · Cue-reward association | validation error | No clear effect |
| Vary the auxiliary output vector ỹ_t (ablation) | Hebbian plasticity · Cue-reward association | validation error | No clear effect |
| Disable plasticity by setting η₀ = 0 (ablation) | Plastic RNNs · Cue-reward association | validation error | Hurt |
Closest papers by meaning
- Differentiable plasticity: training plastic neural networks with backpropagation (ICML 2018 · similarity 0.92)
- Learning the Plasticity: Plasticity-Driven Learning Framework in Spiking Neural Networks (NeurIPS 2025 · similarity 0.92)
- H-Mem: Harnessing synaptic plasticity with Hebbian Memory Networks (NeurIPS 2020 · similarity 0.91)
- Model Based Inference of Synaptic Plasticity Rules (NeurIPS 2024 · similarity 0.90)
- Discovering plasticity rules that organize and maintain neural circuits (NeurIPS 2024 · similarity 0.87)
- Predictive Attractor Models (NeurIPS 2024 · 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.