Network Dissection and Broden: concept-aligned CNN unit analysis

What changed

Thresholds and upsamples convolutional unit activations and compares them with densely annotated concepts through intersection-over-union, enabling representation comparisons and rotation tests.

Scientific scope

Thresholds and upsamples convolutional unit activations and compares them with densely annotated concepts through intersection-over-union, enabling representation comparisons and rotation tests. Interpretability is relative to Broden coverage, thresholds and individual axis-aligned units, not an exhaustive causal decomposition. Units matching absent concepts can be missed. First two authors explicitly equal.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Interpretability is relative to Broden coverage, thresholds and individual axis-aligned units, not an exhaustive causal decomposition. Units matching absent concepts can be missed. First two authors explicitly equal.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Network Dissection: Quantifying Interpretability of Deep Visual Representations

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction