---
aliases:
  en:
  - mech interp
  - mechanistic interp
  - mechinterp
  ja:
  - メカインタープ
concept_id: mechanistic_interpretability
preferred:
  en: mechanistic interpretability
  ja: メカニスティック・インタープリタビリティ
relations:
  relations:
    broader:
    - interpretability
    prerequisite:
    - interpretability
    - neural_network
status: active
---

# mechanistic interpretability

A subfield of interpretability that seeks to explain model behavior by analyzing internal mechanisms such as weights, activations, features, and circuits, especially in neural networks.

## broader

- [interpretability](../terminology/interpretability.txt)

## prerequisite

- [interpretability](../terminology/interpretability.txt)
- [neural network](../terminology/neural_network.txt)

## Documents: about

- [一見 monosemantic な SAE latent でも、本来発火すべき入力を取りこぼすことがある](../documents/chanin_et_al.monosemantic-sae-features-can-have-systematic-false-negatives.txt)
- [SAE の辞書幅を大きくしても、一意・完全・不可分な feature 辞書に収束するとは限らない](../documents/leask_et_al.sae-dictionaries-need-not-be-canonical.txt)

## Documents: mentions

- [解釈可能性は危険の診断に役立っても、それだけでアラインメントの解決策にはならない](../documents/yudkowsky.interpretability-is-diagnostic-not-by-itself-an-alignment-solution.txt)

[Corpus index](../index.txt)