Research map / Label-efficient and robust learning / Self-supervised and contrastive
Scaling Neural Motor Decoding via Decoupled Behavioral Pretraining
Divyansha ., Vinam Arora, Shivashriganesh P. Mahato, Alexandre Andre, Eva L Dyer
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes BeeMO, a neural motor decoder that decouples behavioral and neural representation learning so that abundant behavioral data can be used for pretraining before neural-behavioral alignment.
Problem
Neural motor decoding typically requires large paired neural-behavioral recordings, which are expensive, invasive, and hard to scale.
Approach
Decouple behavioral and neural representation learning: pretrain on abundant behavioral data and then link neural signals to the behavioral representation, rather than training only on paired neural-behavior data.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add ~6.5h unpaired behavioral data to pretraining | POYO · RT cross-task (CO-pretrained) | R2: 0.7498 -> 0.7609 | Improved |
| add ~6.5h unpaired behavioral data to pretraining | POYO · Isometric Wrist cross-task | R2: 0.7237 -> 0.7361 | Improved |
| add ~6.5h unpaired behavioral data to pretraining | NDT* · RT cross-task | R2: 0.7600 -> 0.7734 | Improved |
| add ~6.5h unpaired behavioral data to pretraining | NDT* · Isometric Wrist cross-task | R2: +0.0035 | No clear effect |
| add ~6.5h unpaired behavioral data to pretraining | POYO / NDT* · CO within-task | R2: +0.0058 (POYO), +0.0017 (NDT*) | No clear effect |
| add behavior dropout during training (ablation) | BeeMO-POYO · RT cross-task | R2: 0.6925 -> 0.7609 | Improved |
| add behavior dropout during training (ablation) | BeeMO-NDT* · RT cross-task | R2: 0.7101 -> 0.7734 | Improved |
| add unpaired behavioral data (vs w/o unpaired) (ablation) | BeeMO-POYO · Isometric Wrist cross-task | R2: 0.6849 -> 0.7361 | Improved |
| add unpaired behavioral data (vs w/o unpaired) (ablation) | BeeMO-NDT* · Isometric Wrist cross-task | R2: 0.7757 -> 0.7802 | Improved |
| replace decoder with deeper BeeMO decoder without unpaired data (ablation) | POYO · Isometric Wrist cross-task | R2: -0.0388 R2 | Hurt |
| add task-matched (IW) behavioral data, 2.5h | BeeMO-POYO · IW, held-out animal | R2: approx +0.015 | Improved |
| add task-matched (IW) behavioral data, 2.5h | BeeMO-NDT* · IW, held-out animal | R2: approx +0.0035 | Improved |
Closest papers by meaning
- SPINT: Spatial Permutation-Invariant Neural Transformer for Consistent Intracortical Motor Decoding (NeurIPS 2025 · similarity 0.90)
- Neural Encoding and Decoding at Scale (ICML 2025 · similarity 0.90)
- A Scalable, Causal, and Energy Efficient Framework for Neural Decoding with Spiking Neural Networks (NeurIPS 2025 · similarity 0.89)
- Leveraging unlabelled data for generalizable neural population decoding (NeurIPS 2026 · similarity 0.89)
- Few-shot Algorithms for Consistent Neural Decoding (FALCON) Benchmark (NeurIPS 2024 · similarity 0.88)
- Neural-Behavioral Representation of Natural Whole-body Movement in Monkeys (NeurIPS 2026 · 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.
