Research map / Probabilistic and scientific modeling / Dynamical systems
Learning and processing the ordinal information of temporal sequences in recurrent neural circuits
Xiaolong Zou, Zhikun Chu, Qinghai Guo, Jie Cheng, Bo Hong, Si Wu, Yuanyuan Mi
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 proposed training protocol (interval augmentation + ramping targets) instead of conventional (ablation) | GRU · TIMIT key-word spotting, 4 classes, warped test | accuracy: 88.9% -> 96.6% | Improved |
| use proposed training protocol (interval augmentation + ramping targets) instead of conventional (ablation) | GRU · TIMIT key-word spotting, 8 classes, warped test | accuracy: 87.9% -> 95.6% | Improved |
| use proposed training protocol (interval augmentation + ramping targets) instead of conventional (ablation) | GRU · TIMIT key-word spotting, 12 classes, warped test | accuracy: 84.4% -> 94.8% | Improved |
| use proposed training protocol (interval augmentation + ramping targets) instead of conventional (ablation) | LSTM · TIMIT key-word spotting, 4 classes, warped test | accuracy: 82.8% -> 92.5% | Improved |
| use proposed training protocol (interval augmentation + ramping targets) instead of conventional (ablation) | LSTM · TIMIT key-word spotting, 8 classes, warped test | accuracy: 80.5% -> 90.1% | Improved |
| use proposed training protocol (interval augmentation + ramping targets) instead of conventional (ablation) | LSTM · TIMIT key-word spotting, 12 classes, warped test | accuracy: 75.9% -> 89.5% | Improved |
| use proposed training protocol instead of conventional (ablation) | GRU · TIMIT key-word spotting, 4 classes, no warping | accuracy: 94.0% -> 97.4% | Improved |
| use proposed training protocol instead of conventional (ablation) | LSTM · TIMIT key-word spotting, 12 classes, no warping | accuracy: 92.1% -> 92.6% | No clear effect |
| freeze recurrent weights and reuse stored tree-structured order template (ablation) | RNN with tree attractors · transfer learning, TIMIT phoneme sequences (same order structure) | learning speed | Improved |
| reuse only part of the stored order template (six new sequences) (ablation) | RNN with tree attractors · partial-structure transfer learning | learning speed | Improved |
| use ramping target function instead of cross-entropy loss in pretraining (ablation) | RNN transfer learning · transfer learning | learning speed | Improved |
| train with warped sequences (warp 1 to 1.5) and cross-entropy instead of proposed protocol (control 1) (ablation) | RNN · key-word spotting, warping outside training range | accuracy | Hurt |
Closest papers by meaning
- From sequences to schemas: low-rank recurrent dynamics underlie abstract relational representations (NeurIPS 2026 · similarity 0.90)
- Dynamical phases of short-term memory mechanisms in RNNs (ICML 2025 · similarity 0.86)
- Predictive Attractor Models (NeurIPS 2024 · similarity 0.86)
- Traveling Waves Encode The Recent Past and Enhance Sequence Learning (ICLR 2024 · similarity 0.84)
- Leveraging Attractor Dynamics in Spatial Navigation for Better Language Parsing (ICML 2024 · similarity 0.84)
- Complex priors and flexible inference in recurrent circuits with dendritic nonlinearities (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.
