Research map / Architectures and efficiency / RNNs and normalization
Learning rule influences recurrent network representations but not attractor structure in decision-making tasks
Brandon J McMahan, Michael Kleinman, Jonathan Kao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| train with BPTT, GA or FF instead of Hebbian learning | RNN representations · RDM and CDI | SVCCA MDS overlap | Mixed |
| vary learning rule (BPTT, GA, H, FF) | RNN attractor topology · RDM and CDI | attractor silhouette: -0.05; 0.004 | No clear effect |
| increase number of contexts N from 2 to 6 | BPTT vs GA representations · N-CDI | SVCCA silhouette: 0.10 -> 0.44 | Improved |
| increase number of contexts N from 2 to 6 | BPTT vs GA attractor structure · N-CDI | attractor silhouette: 0.030 -> 7e-4 | No clear effect |
| increase task complexity beyond N=2 | H and FF learning · N-CDI | reaching 90% accuracy | Hurt |
| increase input noise variance | attractor topology · RDM | number of attractors at low input | Improved |
| increase input noise variance | universality across learning rules · RDM | attractor silhouette | Improved |
| add high input noise (variance 1.0 vs 0.1) | integration mechanism · RDM | integration type | Mixed |
| increase N-CDI contexts beyond 3 | input-noise effect on attractor topology (BPTT) · N-CDI | attractor similarity across noise 0.1, 0.5, 1.0 | Improved |
| change activation function (ablation) | attractor topology · RDM vs CDI | attractor topology | Mixed |
| use single neuron readout for non-Hebbian RNNs (ablation) | SVCCA representations · RDM and CDI | SVCCA clustering | No clear effect |
Closest papers by meaning
- Measuring and Controlling Solution Degeneracy across Task-Trained Recurrent Neural Networks (NeurIPS 2025 · similarity 0.89)
- Dynamical phases of short-term memory mechanisms in RNNs (ICML 2025 · similarity 0.88)
- Learning Dynamics of RNNs in Closed-Loop Environments (NeurIPS 2025 · similarity 0.88)
- RNNs perform task computations by dynamically warping neural representations (NeurIPS 2025 · similarity 0.88)
- Feedback control guides credit assignment in recurrent neural networks (NeurIPS 2024 · similarity 0.88)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.88)
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.
