Research map / Architectures and efficiency / RNNs and normalization
Balancing memorization and generalization in RNNs for high performance brain-machine Interfaces
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 |
|---|---|---|---|
| use LSTM instead of other decoders (GRU, FNN, TFM, KF) | finger BMI decoder · Offline 2-DoF random task | MSE | Helped |
| use LSTM instead of FNN online | finger BMI decoder · Online finger task, Monkey N | median bitrate: 18% higher than FNN | Helped |
| use LSTM instead of GRU online | finger BMI decoder · Online finger task, Monkey N | median bitrate: 15% higher than GRU | Helped |
| increase number of simulated channels from 10-100 to 400 | LSTM vs TFM · Simulated offline decoding | correlation gap: 0.1 -> 0.01 gap | Hurt |
| restrict training to few targets so the RNN can memorize movements | LSTM decoder · Online 2-4 target tasks | bitrate vs hand control: +2% vs hand control | Helped |
| increase number of targets in training | LSTM decoder · Offline target tasks | MSE | Hurt |
| train with only three index finger targets while keeping random MRS targets | LSTM decoder with degraded index signal · Online 2-DoF random task | success rate: 66% -> 98% | Helped |
| use more time history (up to 20 bins) in FNN and TFM (ablation) | FNN and TFM decoders · Online control | online control quality | Hurt |
| switch memorized decoder from positional to velocity variant for generalizing to random targets (ablation) | 2-target LSTM · Online random target task | success rate: <50% -> 98% | Helped |
| modify loss to encourage finger independence (ablation) | online decoders · Online finger task (early exploration) | success rate: 11% increase in SR | Helped |
| train on more single-finger movements (ablation) | online decoders · Online finger task (early exploration) | success rate: 38% increase in SR | Helped |
| add noise to neural data during training (ablation) | online decoders · Online finger task | trial time: 6% | Helped |
Closest papers by meaning
- Exploring the trade-off between deep-learning and explainable models for brain-machine interfaces (NeurIPS 2024 · similarity 0.90)
- SPINT: Spatial Permutation-Invariant Neural Transformer for Consistent Intracortical Motor Decoding (NeurIPS 2025 · similarity 0.86)
- Few-shot Algorithms for Consistent Neural Decoding (FALCON) Benchmark (NeurIPS 2024 · similarity 0.86)
- Generalizable, real-time neural decoding with hybrid state-space models (NeurIPS 2025 · similarity 0.86)
- A Scalable, Causal, and Energy Efficient Framework for Neural Decoding with Spiking Neural Networks (NeurIPS 2025 · similarity 0.86)
- Learning Dynamics of RNNs in Closed-Loop Environments (NeurIPS 2025 · similarity 0.85)
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.