Towards a fuller understanding of neurons with Clustered Compositional Explanations

Biagio La Rosa, Leilani Gilpin, Roberto Capobianco
10/1/2023

No official code found. Best implementation: KRLGroup/Clustered-Compositional-Explanations (6 stars, MIT, last commit 2025-10-21).

Abstract

Compositional Explanations is a method for identifying logical formulas of concepts that approximate the neurons' behavior. However, these explanations are linked to the small spectrum of neuron activations (i.e., the highest ones) used to check the alignment, thus lacking completeness. In this paper, we propose a generalization, called Clustered Compositional Explanations, that combines Compositional Explanations with clustering and a novel search heuristic to approximate a broader spectrum of the neurons' behavior. We define and address the problems connected to the application of these methods to multiple ranges of activations, analyze the insights retrievable by using our algorithm, and propose desiderata qualities that can be used to study the explanations returned by different algorithms.

View on arXivView PDF

Code Implementations(1)

KRLGroup/Clustered-Compositional-ExplanationsCommunity63%
MIT

Official repository of the paper "Towards a fuller understanding of neurons with Clustered Compositional Explanations. Biagio La Rosa, Leilani Gilpin, and Roberto Capobianco. NeurIPS 2023"

63Shell, PythonOct 19, 20231 years agoMIT

Similar Papers

Guaranteed Optimal Compositional Explanations for Neurons

90% match1 repo32025

Open Vocabulary Compositional Explanations for Neuron Alignment

86% match1 repo2025

Dynamic MOdularized Reasoning for Compositional Structured Explanation Generation

80% match2023

NeurFlow: Interpreting Neural Networks through Neuron Groups and Functional Interactions

80% match1 repo72025

On the Value of Labeled Data and Symbolic Methods for Hidden Neuron Activation Analysis

79% match2024

Linear Explanations for Individual Neurons

78% match1 repo142024

Cite this paper

@article{rosa2023towards,
  title  = {Towards a fuller understanding of neurons with Clustered Compositional Explanations},
  author = {Biagio La Rosa and Leilani Gilpin and Roberto Capobianco},
  year   = {2023},
  eprint = {2310.18443},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/2310.18443},
  journal = {NEURIPS 2023 2023}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2310.18443.svg)](https://codeofpaper.com/paper/2310.18443)

Discussion