Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift

• URL: http://arxiv.org/abs/2605.21849v1
• Date: Thu, 21 May 2026 00:46:01 GMT
• Title: Geometry-Adaptive Explainer for Faithful Dictionary-Based Interpretability under Distribution Shift
• Abstract summary: We show that distribution shift rotates the subspace that the model actively uses, misaligning the explainer's dictionary trained on in-distribution (ID) activations.<n>We formalize this misalignment as the faithfulness gap, a geometric distance between the ID dictionary and the OOD-active subspace.<n>We propose the Geometry-Adaptive Explainer (GAE), which realigns the explainer's dictionary with the OOD-active subspace while preserving the original feature structure.
• Score: 17.611062308867275
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Mechanistic interpretability aims to explain a model's behavior by identifying causally responsible internal structures. Dictionary-based explainers such as sparse autoencoders and transcoders are a primary tool, but their faithfulness under out-of-distribution (OOD) shift has received little systematic attention. We show that distribution shift rotates the subspace that the model actively uses, misaligning the explainer's dictionary trained on in-distribution (ID) activations. We formalize this misalignment as the faithfulness gap, a geometric distance between the ID dictionary and the OOD-active subspace, and show that it controls OOD faithfulness degradation. To reduce this gap, we propose the Geometry-Adaptive Explainer (GAE), which realigns the explainer's dictionary with the OOD-active subspace while preserving the original feature structure. This requires only unlabeled OOD activations and no gradient updates. We prove that GAE improves over the unadapted ID explainer, with excess loss bounded quadratically by the second-moment shift. Empirically, GAE even matches or surpasses all training-based baselines in causal faithfulness across multiple models and OOD settings.

Related papers

• DriftScope: Measuring The Hidden Effects of Diffusion Model Adaptation [50.584834483403675]
  Adapting pre-trained text-to-image diffusion models, whether to learn new visual concepts or erase unwanted ones, is routinely evaluated on its intended effects alone.<n>We demonstrate that adaptation systematically damages semantically unrelated concepts in ways that aggregate metrics structurally cannot surface.<n>We introduce DriftScope, a prompt-level diagnostic tool that takes any two model checkpoints and returns a ranked list of tokens whose visual concepts have shifted most between them.
  arXiv  Detail & Related papers  (2026-06-30T20:57:51Z)
• MENTIS: What Belief Changes Under Alignment? Measuring Multi-Scale Latent Torsion in Language Models [16.071341117554038]
  We introduce MENTIS, a framework for measuring alignment-induced internal reorganization.<n>Across four 7-8B model pairs on LITMUS, our study reveals that alignment-induced change is selective rather than uniform.<n>These results suggest preference alignment leaves structured, depth-localized geometric signatures in internal computation beyond what behavior-level evaluation alone can reveal.
  arXiv  Detail & Related papers  (2026-05-31T07:05:51Z)
• Is Inference Mediated by Distinct Semantic Structures in LLMs? A Mechanistic Interpretation [34.50304918655722]
  Transformer representations are known to carry label-level information, but whether they encode semantic operations producing those labels is unclear.<n>We investigate this in Natural Language Inference using controlled premise-hypothesis pairs that differ by a single semantic transformation.
  arXiv  Detail & Related papers  (2026-05-25T07:21:07Z)
• TAPIOCA: Why Task- Aware Pruning Improves OOD model Capability [4.56250586729666]
  Recent work has promoted task-aware layer pruning as a way to improve model performance on particular tasks.<n>We show that such pruning yields no benefit on in-distribution data but consistently improves out-of-distribution accuracy.<n>This leads to a geometric explanation of task-aware pruning.
  arXiv  Detail & Related papers  (2026-05-14T12:01:05Z)
• Self-Corrected Image Generation with Explainable Latent Rewards [55.29175717238288]
  We propose xLARD, a self-correcting framework that guides generation through Explainable LAtent RewarDs.<n>xLARD introduces a lightweight corrector that refines latent representations based on structured feedback from model-generated references.<n> Experiments show that xLARD improves semantic alignment and visual fidelity while maintaining generative priors.
  arXiv  Detail & Related papers  (2026-03-26T02:59:35Z)
• A Signed Graph Approach to Understanding and Mitigating Oversmoothing in GNNs [54.62268052283014]
  We present a unified theoretical perspective based on the framework of signed graphs.<n>We show that many existing strategies implicitly introduce negative edges that alter message-passing to resist oversmoothing.<n>We propose Structural Balanced Propagation (SBP), a plug-and-play method that assigns signed edges based on either labels or feature similarity.
  arXiv  Detail & Related papers  (2025-02-17T03:25:36Z)
• Semantic or Covariate? A Study on the Intractable Case of Out-of-Distribution Detection [70.57120710151105]
  We provide a more precise definition of the Semantic Space for the ID distribution. We also define the "Tractable OOD" setting which ensures the distinguishability of OOD and ID distributions.
  arXiv  Detail & Related papers  (2024-11-18T03:09:39Z)
• Dissecting the Failure of Invariant Learning on Graphs [36.11431280689549]
  We develop a Structural Causal Model (SCM) to theoretically dissect the performance of two prominent invariant learning methods.<n>We propose Cross-environment Intra-class Alignment (CIA), which explicitly eliminates spurious features by aligning cross-environment representations conditioned on the same class.<n>We further propose CIA-LRA (Localized Reweighting Alignment) that leverages the distribution of neighboring labels to selectively align node representations.
  arXiv  Detail & Related papers  (2024-11-05T06:36:48Z)
• Diagnosing and Rectifying Fake OOD Invariance: A Restructured Causal Approach [51.012396632595554]
  Invariant representation learning (IRL) encourages the prediction from invariant causal features to labels de-confounded from the environments. Recent theoretical results verified that some causal features recovered by IRLs merely pretend domain-invariantly in the training environments but fail in unseen domains. We develop an approach based on conditional mutual information with respect to RS-SCM, then rigorously rectify the spurious and fake invariant effects.
  arXiv  Detail & Related papers  (2023-12-15T12:58:05Z)
• Disentanglement via Latent Quantization [60.37109712033694]
  In this work, we construct an inductive bias towards encoding to and decoding from an organized latent space. We demonstrate the broad applicability of this approach by adding it to both basic data-re (vanilla autoencoder) and latent-reconstructing (InfoGAN) generative models.
  arXiv  Detail & Related papers  (2023-05-28T06:30:29Z)

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.