Research map / Architectures and efficiency / RNNs and normalization
Using Conceptors to Manage Neural Long-Term Memories for Temporal Patterns
Herbert Jaeger
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 |
|---|---|---|---|
| auto-adapt the conceptor after cueing instead of using the preliminary cue conceptor (ablation) | Autoconceptor recall · 5-periodic and mix-of-sines patterns | NRMSE | Improved |
| add state noise during autonomous auto-adaptation (ablation) | Autoconceptor recall, mix of sines · Mix of 2 sines (10 patterns) | adaptation steps needed: 1,000,000 steps without noise vs 10,000 with noise | Improved |
| remove cue and state noise (ablation) | Autoconceptor recall, 5-periodic patterns · 5-periodic patterns | NRMSE: 0.070 -> 0.0037 | Improved |
| use autoconceptors instead of stored alloconceptors | Conceptor recall, mix of sines · Mix of 2 sines (10 patterns) | NRMSE: 0.087 -> 0.11 | Hurt |
| use autoconceptors instead of stored alloconceptors under noise | Conceptor recall, 5-periodic patterns · 5-periodic patterns | NRMSE: .0033 -> 0.070 | Hurt |
| adapt conceptor with loss on filtered state z instead of reservoir output r (ablation) | Autoconceptor adaptation · 5-periodic patterns | noise robustness and singular value profile | Improved |
| load more patterns from a parametric family (ablation) | Autoconceptor recall of novel patterns · Mix of 2 sines family, N=100 | log10 NRMSE of novel patterns: within 20% of loaded-pattern accuracy at about 5 patterns | Improved |
| load more patterns into the reservoir (ablation) | Autoconceptor recall of loaded patterns · Mix of 2 sines family, N=100 | NRMSE of loaded patterns | Hurt |
| set aperture far from the optimal value (ablation) | Alloconceptor recall of Lorenz attractor · Lorenz attractor | visual match | No clear effect |
Closest papers by meaning
- Predictive Attractor Models (NeurIPS 2024 · similarity 0.87)
- Dynamical phases of short-term memory mechanisms in RNNs (ICML 2025 · similarity 0.86)
- Continuous Thought Machines (NeurIPS 2025 · similarity 0.85)
- Back to the Continuous Attractor (NeurIPS 2024 · similarity 0.84)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.84)
- Emergent mechanisms for long timescales depend on training curriculum and affect performance in memory tasks (ICLR 2024 · similarity 0.84)
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.
