科研 Skill

脑机解码器长期稳定性

neuro-ibci-decoder-stability

Evaluates intracortical BCI decoders across days using zero-shot and few-shot regimes to quantify drift and recalibration cost. Use to assess decoder stability over sessions.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

Intracortical decoder stability across days

Signals from chronic implants drift: units appear and disappear, thresholds shift, and neural tuning changes. A decoder scored within one session says nothing about tomorrow. Evaluate in time order and report how much recalibration data each method needs.

1. Organise data by day

Index every trial by session date. Never shuffle across days. Keep the features fixed in definition (same channels, bin width, threshold rule) so later days are comparable; log any channel that is dead or re-referenced.

2. Use three evaluation regimes

RegimeTrainTestAnswers
Within-dayearly part of day dlater part of day dceiling
Zero-shotdays 1..dday d+k, no new datahow fast does it decay
Few-shotdays 1..d + first m trials of day d+krest of day d+krecalibration cost

This matches the FALCON benchmark design (calibration, few-shot and held-out evaluation days; Karpowicz et al. 2024). Report performance against days since training, and against m (minutes or trials of recalibration).

3. Metrics

  • Continuous decoding (cursor, kinematics): R^2 or correlation per dimension on held-out time, not on shuffled bins.
  • Discrete (keys, phonemes, words): accuracy plus error rate per unit time.
  • Closed-loop matters more than offline: if online data exist, report bit rate or task success, because offline R^2 can rank methods wrongly.

4. Baselines every comparison needs

  1. Fixed decoder trained on day 1 only (lower bound).
  2. Daily full recalibration (upper bound for cost).
  3. Simple mean/variance re-normalisation per day (cheap, often strong).
  4. Your method. Show it beats 3, not only 1.

5. Diagnose why performance drops

Plot per-channel firing-rate change, and decoder weights on channels that changed. Separate recording change (lost units, impedance) from neural change (tuning shift). Recording change argues for channel-robust features; neural change argues for adaptive decoders.

Report

Days spanned, sessions, channels kept, regimes above with performance vs days and vs recalibration data, the three baselines, and online results if any. Note the participant count: results from one participant are a case study.

示例提示词

帮我评估这个皮层内BCI解码器跨天的漂移情况和重标定方案。

所属插件

神经科学面向神经科学研究:EEG/MEG、fMRI、BIDS、spike sorting、连接组、脑机接口解码评估、神经刺激安全与公开神经数据检索。

安装该插件会一并获得同领域的 45 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

脑机解码器长期稳定性能做什么?

Evaluates intracortical BCI decoders across days using zero-shot and few-shot regimes to quantify drift and recalibration cost. Use to assess decoder stability over sessions。

如何安装脑机解码器长期稳定性?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「脑机解码器长期稳定性」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

脑机解码器长期稳定性需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

脑机解码器长期稳定性经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。