---
source: https://iinventedlight.com/library/notes/ai-feature-visualization-attribution.md
title: "Feature visualization and attribution answer different questions"
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
source_license: CC-BY-4.0
attribution: "Adapted summaries of the named authors below; wording and synthesis by Librarian."
author: Librarian
date: 2026-11-05
status: published
checked_on: 2026-11-05
as_of: 2026-11-05
confidence: high
untested: "Literature summary; no model experiments run."
category: coverage-gaps
topic_area: "ai-research: evaluation and interpretability"
tags: [ai-research, research-methods]
sources:
  - id: S1
    title: "Feature Visualization"
    publisher: "Distill"
    url: https://distill.pub/2017/feature-visualization/
    accessed: 2026-11-05
    license: CC-BY-4.0
    attribution: "Olah, Chris; Mordvintsev, Alexander; Schubert, Ludwig"
  - id: S2
    title: "The Building Blocks of Interpretability"
    publisher: "Distill"
    url: https://distill.pub/2018/building-blocks/
    accessed: 2026-11-05
    license: CC-BY-4.0
    attribution: "Olah, Chris; Satyanarayan, Arvind; Johnson, Ian; Carter, Shan; Schubert, Ludwig; Ye, Katherine; Mordvintsev, Alexander"
---
# Feature visualization and attribution answer different questions

Feature visualization generates examples of what a network or part of a network is looking for. [S1]

## Key distinctions

Attribution concerns which parts of an example are responsible for a network activating. [S1]

Optimization can adjust an input to activate a particular neuron, starting from random noise. [S1]

Interpretability interfaces can combine methods that have normally been studied in isolation. [S2]

## A useful distinction

Feature visualization helps answer what the network detects. [S2]

It does not by itself answer how the network assembles the pieces into later decisions. [S2]

## Claims

- id: C1
  claim: "Feature visualization generates examples of what a network or part of a network is looking for."
  source: S1
  evidence: "Feature visualization answers questions about what a network"

- id: C2
  claim: "Attribution concerns which parts of an example are responsible for a network activating."
  source: S1
  evidence: "what part of an example is responsible for the network activating a particular way"

- id: C3
  claim: "Optimization can adjust an input to activate a particular neuron, starting from random noise."
  source: S1
  evidence: "Starting from random noise, we optimize an image to activate a particular neuron"

- id: C4
  claim: "Interpretability interfaces can combine methods that have normally been studied in isolation."
  source: S2
  evidence: "Interpretability techniques are normally studied in isolation."

## Sources

- [S1]

Feature Visualization. Distill. https://distill.pub/2017/feature-visualization/ CC BY 4.0; accessed 2026-11-05.
- [S2]

The Building Blocks of Interpretability. Distill. https://distill.pub/2018/building-blocks/ CC BY 4.0; accessed 2026-11-05.

Attribution: source authors named above; adapted summaries, not endorsed by the authors.