Research map / Continual learning / Online and replay-based
Operative dimensions in unconstrained connectivity of recurrent neural networks
Renate Barbara Krause, Matthew Cook, Sepp Kollmorgen, Valerio Mante, Giacomo Indiveri
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 |
|---|---|---|---|
| Use operative dimensions instead of high-variance weight PCs to build reduced-rank W | Vanilla RNN, context-dependent integration · Context-dependent integration | dimensions needed for original performance: 88 -> 15 | Improved |
| Use operative dimensions instead of high-variance weight PCs to build reduced-rank W | Vanilla RNN, sine wave generation · Sine wave generation | dimensions needed for original performance: 86 -> 29 | Improved |
| Define operative dimensions from random sampling locations rather than along condition-average trajectories (ablation) | Operative dimensions · Context-dependent integration and sine wave generation | network output cost at fixed rank | Hurt |
| Use too few sampling locations along trajectories (ablation) | Operative dimensions · Context-dependent integration and sine wave generation | reproduction of network output and trajectories | Hurt |
| Add first few operative dimensions to weight matrix | Operative-dimension reduced-rank RNN · Context-dependent integration and sine wave generation | network output cost | Hurt |
| Increase number of input and output signals from 1 to 5 | Sine wave generator RNN · Multi-sine wave generation | operative dimensions needed | Mixed |
| Use operative dimensions from context 1 to run context 2 (and vice versa) | Context-dependent integrator · Context-dependent integration | network output cost | Hurt |
| Use operative dimensions from one choice to run the other choice | Context-dependent integrator · Context-dependent integration | network output cost | No clear effect |
| Increase rank k of high-variance reduced W | Vanilla RNN · Context-dependent integration | network output cost: k=60 to k=61 | Mixed |
Closest papers by meaning
- Learning Reveals Invisible Structure in Low-Rank RNNs (NeurIPS 2026 · similarity 0.89)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.89)
- How connectivity structure shapes rich and lazy learning in neural circuits (ICLR 2024 · 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 0.87)
- Structured flexibility in recurrent neural networks via neuromodulation (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.
