From Mechanistic to Compositional Interpretability

• URL: http://arxiv.org/abs/2605.08934v1
• Date: Sat, 09 May 2026 13:08:07 GMT
• Title: From Mechanistic to Compositional Interpretability
• Abstract summary: Mechanistic interpretability aims to explain neural model behaviour by reverse-engineering learned computational structure into human-understandable components.<n>We introduce compositional interpretability, a category-theoretic framework grounded in the principles of compositionality and minimum description length.
• Score: 1.4625665070654144
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Mechanistic interpretability aims to explain neural model behaviour by reverse-engineering learned computational structure into human-understandable components. Without a formal framework, however, mechanistic explanations cannot be objectively verified, compared, or composed. We introduce compositional interpretability, a category-theoretic framework grounded in the principles of compositionality and minimum description length. Compositional interpretations are pairs of syntactic and semantic mappings that must commute to enforce consistency between a model's decomposition and its observed behaviour. We deconstruct explanation quality into measures of faithfulness and complexity to cast interpretability as a constrained optimisation problem, and introduce compressive refinement to systematically restructure models into simpler parts without altering their function. Finally, we prove a parsimony criterion under which syntactic compression theoretically guarantees more concise, human-aligned explanations. Our framework situates prominent mechanistic methods as subclasses of refinement, and clarifies why their compressibility heuristics tend to align with human interpretability. Our work provides a measurable, optimisable foundation for automating the discovery and evaluation of mechanistic explanations.

Related papers

• Loss-Complexity Landscape and Model Structure Functions [53.92822954974537]
  We develop a framework for dualizing the Kolmogorov structure function $h_x(alpha)$.<n>We establish a mathematical analogy between information-theoretic constructs and statistical mechanics.<n>We explicitly prove the Legendre-Fenchel duality between the structure function and free energy.
  arXiv  Detail & Related papers  (2025-07-17T21:31:45Z)
• Mechanisms vs. Outcomes: Probing for Syntax Fails to Explain Performance on Targeted Syntactic Evaluations [31.54093034816165]
  Large Language Models (LLMs) exhibit a robust mastery of syntax when processing and generating text.<n>No comprehensive study has yet established whether a model's probing accuracy reliably predicts its downstream syntactic performance.
  arXiv  Detail & Related papers  (2025-06-20T01:46:50Z)
• How do Transformers Learn Implicit Reasoning? [67.02072851088637]
  We study how implicit multi-hop reasoning emerges by training transformers from scratch in a controlled symbolic environment.<n>We find that training with atomic triples is not necessary but accelerates learning, and that second-hop generalization relies on query-level exposure to specific compositional structures.
  arXiv  Detail & Related papers  (2025-05-29T17:02:49Z)
• A Mathematical Philosophy of Explanations in Mechanistic Interpretability -- The Strange Science Part I.i [0.0]
  We argue that Mechanistic Interpretability research is a principled approach to understanding models.<n>We show that Explanatory Faithfulness, an assessment of how well an explanation fits a model, is well-defined.
  arXiv  Detail & Related papers  (2025-05-01T19:08:34Z)
• Validating Mechanistic Interpretations: An Axiomatic Approach [28.336108192282737]
  We give a set of axioms that formally characterize a mechanistic interpretation as a description that approximately captures the semantics of the neural network under analysis in a compositional manner.<n>We demonstrate the applicability of these axioms for validating mechanistic interpretations on an existing, well-known interpretability study as well as on a new case study involving a Transformer-based model trained to solve the well-known 2-SAT problem.
  arXiv  Detail & Related papers  (2024-07-18T15:32:44Z)
• What makes Models Compositional? A Theoretical View: With Supplement [60.284698521569936]
  We propose a general neuro-symbolic definition of compositional functions and their compositional complexity. We show how various existing general and special purpose sequence processing models fit this definition and use it to analyze their compositional complexity.
  arXiv  Detail & Related papers  (2024-05-02T20:10:27Z)
• Multi-Relational Hyperbolic Word Embeddings from Natural Language Definitions [5.763375492057694]
  This paper presents a multi-relational model that explicitly leverages such a structure to derive word embeddings from definitions. An empirical analysis demonstrates that the framework can help imposing the desired structural constraints. Experiments reveal the superiority of the Hyperbolic word embeddings over the Euclidean counterparts.
  arXiv  Detail & Related papers  (2023-05-12T08:16:06Z)
• Causal Abstraction: A Theoretical Foundation for Mechanistic Interpretability [30.76910454663951]
  Causal abstraction provides a theoretical foundation for mechanistic interpretability.<n>Our contributions are generalizing the theory of causal abstraction from mechanism replacement to arbitrary mechanism transformation.
  arXiv  Detail & Related papers  (2023-01-11T20:42:41Z)
• Sparse Relational Reasoning with Object-Centric Representations [78.83747601814669]
  We investigate the composability of soft-rules learned by relational neural architectures when operating over object-centric representations. We find that increasing sparsity, especially on features, improves the performance of some models and leads to simpler relations.
  arXiv  Detail & Related papers  (2022-07-15T14:57:33Z)
• Compositional Processing Emerges in Neural Networks Solving Math Problems [100.80518350845668]
  Recent progress in artificial neural networks has shown that when large models are trained on enough linguistic data, grammatical structure emerges in their representations. We extend this work to the domain of mathematical reasoning, where it is possible to formulate precise hypotheses about how meanings should be composed. Our work shows that neural networks are not only able to infer something about the structured relationships implicit in their training data, but can also deploy this knowledge to guide the composition of individual meanings into composite wholes.
  arXiv  Detail & Related papers  (2021-05-19T07:24:42Z)
• Evaluations and Methods for Explanation through Robustness Analysis [117.7235152610957]
  We establish a novel set of evaluation criteria for such feature based explanations by analysis. We obtain new explanations that are loosely necessary and sufficient for a prediction. We extend the explanation to extract the set of features that would move the current prediction to a target class.
  arXiv  Detail & Related papers  (2020-05-31T05:52:05Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.