Interpretability Illusions with Sparse Autoencoders: Evaluating Robustness of Concept Representations

• URL: http://arxiv.org/abs/2505.16004v1
• Date: Wed, 21 May 2025 20:42:05 GMT
• Title: Interpretability Illusions with Sparse Autoencoders: Evaluating Robustness of Concept Representations
• Authors: Aaron J. Li, Suraj Srinivas, Usha Bhalla, Himabindu Lakkaraju,
• Abstract summary: We develop an evaluation framework featuring realistic scenarios in which adversarial perturbations are crafted to manipulate SAE representations.<n>We find that tiny adversarial input perturbations can effectively manipulate concept-based interpretations in most scenarios.<n>Overall, our results suggest that SAE concept representations are fragile and may be ill-suited for applications in model monitoring and oversight.
• Score: 23.993903128858832
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Sparse autoencoders (SAEs) are commonly used to interpret the internal activations of large language models (LLMs) by mapping them to human-interpretable concept representations. While existing evaluations of SAEs focus on metrics such as the reconstruction-sparsity tradeoff, human (auto-)interpretability, and feature disentanglement, they overlook a critical aspect: the robustness of concept representations to input perturbations. We argue that robustness must be a fundamental consideration for concept representations, reflecting the fidelity of concept labeling. To this end, we formulate robustness quantification as input-space optimization problems and develop a comprehensive evaluation framework featuring realistic scenarios in which adversarial perturbations are crafted to manipulate SAE representations. Empirically, we find that tiny adversarial input perturbations can effectively manipulate concept-based interpretations in most scenarios without notably affecting the outputs of the base LLMs themselves. Overall, our results suggest that SAE concept representations are fragile and may be ill-suited for applications in model monitoring and oversight.

Related papers

• Interpretable Few-Shot Image Classification via Prototypical Concept-Guided Mixture of LoRA Experts [79.18608192761512]
  Self-Explainable Models (SEMs) rely on Prototypical Concept Learning (PCL) to enable their visual recognition processes more interpretable.<n>We propose a Few-Shot Prototypical Concept Classification framework that mitigates two key challenges under low-data regimes: parametric imbalance and representation misalignment.<n>Our approach consistently outperforms existing SEMs by a notable margin, with 4.2%-8.7% relative gains in 5-way 5-shot classification.
  arXiv  Detail & Related papers  (2025-06-05T06:39:43Z)
• Example-Based Concept Analysis Framework for Deep Weather Forecast Models [25.56878415414591]
  We develop an example-based concept analysis framework, which identifies cases that follow a similar inference process as the target instance in a target model.<n>Our framework provides the users with visually and conceptually analogous examples, including the probability of concept assignment to resolve ambiguities in weather mechanisms.
  arXiv  Detail & Related papers  (2025-04-01T14:22:41Z)
• I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data? [76.15163242945813]
  Large language models (LLMs) have led many to conclude that they exhibit a form of intelligence.<n>We introduce a novel generative model that generates tokens on the basis of human-interpretable concepts represented as latent discrete variables.
  arXiv  Detail & Related papers  (2025-03-12T01:21:17Z)
• Concept Layers: Enhancing Interpretability and Intervenability via LLM Conceptualization [2.163881720692685]
  We introduce a new methodology for incorporating interpretability and intervenability into an existing model by integrating Concept Layers into its architecture.<n>Our approach projects the model's internal vector representations into a conceptual, explainable vector space before reconstructing and feeding them back into the model.<n>We evaluate CLs across multiple tasks, demonstrating that they maintain the original model's performance and agreement while enabling meaningful interventions.
  arXiv  Detail & Related papers  (2025-02-19T11:10:19Z)
• Towards Robust and Reliable Concept Representations: Reliability-Enhanced Concept Embedding Model [22.865870813626316]
  Concept Bottleneck Models (CBMs) aim to enhance interpretability by predicting human-understandable concepts as intermediates for decision-making.<n>Two inherent issues contribute to concept unreliability: sensitivity to concept-irrelevant features and lack of semantic consistency for the same concept across different samples.<n>We propose the Reliability-Enhanced Concept Embedding Model (RECEM), which introduces a two-fold strategy: Concept-Level Disentanglement to separate irrelevant features from concept-relevant information and a Concept Mixup mechanism to ensure semantic alignment across samples.
  arXiv  Detail & Related papers  (2025-02-03T09:29:39Z)
• Concept-Based Explainable Artificial Intelligence: Metrics and Benchmarks [0.0]
  Concept-based explanation methods aim to improve the interpretability of machine learning models.<n>We propose three metrics: the concept global importance metric, the concept existence metric, and the concept location metric.<n>We demonstrate that, in many cases, even the most important concepts determined by post-hoc CBMs are not present in input images.
  arXiv  Detail & Related papers  (2025-01-31T16:32:36Z)
• On the Fairness, Diversity and Reliability of Text-to-Image Generative Models [49.60774626839712]
  multimodal generative models have sparked critical discussions on their fairness, reliability, and potential for misuse. We propose an evaluation framework designed to assess model reliability through their responses to perturbations in the embedding space. Our method lays the groundwork for detecting unreliable, bias-injected models and retrieval of bias provenance.
  arXiv  Detail & Related papers  (2024-11-21T09:46:55Z)
• Self-supervised Interpretable Concept-based Models for Text Classification [9.340843984411137]
  This paper proposes a self-supervised Interpretable Concept Embedding Models (ICEMs) We leverage the generalization abilities of Large-Language Models to predict the concepts labels in a self-supervised way. ICEMs can be trained in a self-supervised way achieving similar performance to fully supervised concept-based models and end-to-end black-box ones.
  arXiv  Detail & Related papers  (2024-06-20T14:04:53Z)
• Improving Intervention Efficacy via Concept Realignment in Concept Bottleneck Models [57.86303579812877]
  Concept Bottleneck Models (CBMs) ground image classification on human-understandable concepts to allow for interpretable model decisions. Existing approaches often require numerous human interventions per image to achieve strong performances. We introduce a trainable concept realignment intervention module, which leverages concept relations to realign concept assignments post-intervention.
  arXiv  Detail & Related papers  (2024-05-02T17:59:01Z)
• Sparsity-Guided Holistic Explanation for LLMs with Interpretable Inference-Time Intervention [53.896974148579346]
  Large Language Models (LLMs) have achieved unprecedented breakthroughs in various natural language processing domains. The enigmatic black-box'' nature of LLMs remains a significant challenge for interpretability, hampering transparent and accountable applications. We propose a novel methodology anchored in sparsity-guided techniques, aiming to provide a holistic interpretation of LLMs.
  arXiv  Detail & Related papers  (2023-12-22T19:55:58Z)
• Interpreting Pretrained Language Models via Concept Bottlenecks [55.47515772358389]
  Pretrained language models (PLMs) have made significant strides in various natural language processing tasks. The lack of interpretability due to their black-box'' nature poses challenges for responsible implementation. We propose a novel approach to interpreting PLMs by employing high-level, meaningful concepts that are easily understandable for humans.
  arXiv  Detail & Related papers  (2023-11-08T20:41:18Z)
• GlanceNets: Interpretabile, Leak-proof Concept-based Models [23.7625973884849]
  Concept-based models (CBMs) combine high-performance and interpretability by acquiring and reasoning with a vocabulary of high-level concepts. We provide a clear definition of interpretability in terms of alignment between the model's representation and an underlying data generation process. We introduce GlanceNets, a new CBM that exploits techniques from disentangled representation learning and open-set recognition to achieve alignment.
  arXiv  Detail & Related papers  (2022-05-31T08:53:53Z)

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.