ICA Lens: Interpreting Language Models Without Training Another Dictionary

• URL: http://arxiv.org/abs/2606.11722v1
• Date: Wed, 10 Jun 2026 06:53:58 GMT
• Title: ICA Lens: Interpreting Language Models Without Training Another Dictionary
• Abstract summary: Independent component analysis (ICA) is a classical method for finding non-Gaussian directions in language-model representations.<n>We introduce ICALens, the first practical workflow for stable, efficient, and auditable ICA analysis of language-model representations.<n>ICA is competitive with public SAEs in sparse probing and outperforms them in targeted probe under small-to-medium budgets.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Finding interpretable directions in language-model representations is critical for understanding and controlling model behavior. Sparse autoencoders (SAEs) have become the standard tool for this purpose, but using them as the default first lens often requires training, storing, and evaluating large overcomplete dictionaries. This bottleneck limits rapid exploration and raises a fundamental question: how much interpretable structure is already visible from activation geometry before training another neural dictionary? Our intuition is simple: many interpretable directions are selective on tokens, and these directions should look less Gaussian than random directions. We therefore revisit independent component analysis (ICA), a classical method for finding non-Gaussian directions, as a compact lens for language-model interpretability. We find that ICA has been underestimated for LLM interpretability, because prior uses often relied on off-the-shelf ICA implementations that are brittle on LLM activations and lacked systematic tools for inspecting and evaluating the recovered directions. To bridge these gaps, we introduce ICALens, the first practical workflow for stable, efficient, and auditable ICA analysis of LLM representations. It combines an optimized GPU-parallel FastICA pipeline with LLM-specific stability recipes and better fitting diagnostics, enabling efficient and reliable layer-wise analysis. Across GPT-2 Small, Gemma 2 2B, and Qwen 3.5 2B Base, ICALens efficiently recovers compact, human-interpretable directions without per-layer gradient-based dictionary training. On SAEBench, ICA is competitive with public SAEs in sparse probing and outperforms them in targeted probe perturbation under small-to-medium budgets. These results suggest that ICA should not be viewed as a weak baseline, but as an efficient and complementary first lens for exploring language-model representations.

Related papers

• The Regularizing Power of Language-Training Deepfake Detectors [57.529452351298495]
  Deepfake detection generally suffers from overfitting to low-level domain-specific artifacts.<n>We use a dual-encoder architecture, pairing a frozen specialist detector with a LoRA-tuned MLLM encoder.<n>We employ a reinforcement learning stage that encourages the model to generate descriptive reasoning before classifying.
  arXiv  Detail & Related papers  (2026-05-29T12:01:17Z)
• Language Bias in LVLMs: From In-Depth Analysis to Simple and Effective Mitigation [3.6330067010404705]
  Large Vision-Language Models (LVLMs) extend large language models with visual understanding, but remain vulnerable to hallucination.<n>Recent studies link this issue to language bias-the tendency of LVLMs to over-rely on text while neglecting visual inputs.<n>We provide a systematic study of language bias and identify its root in modality misalignment during training.
  arXiv  Detail & Related papers  (2026-05-24T12:23:13Z)
• Heuristic-inspired Reasoning Priors Facilitate Data-Efficient Referring Object Detection [53.988759250627425]
  HeROD (Heuristic-inspired ROD) is a lightweight, model-agnostic framework that injects explicit, interpretable spatial and semantic reasonings.<n>HeROD consistently outperforms strong grounding baselines in scarcelabel regimes.
  arXiv  Detail & Related papers  (2026-03-25T10:33:22Z)
• Language Bottleneck Models: A Framework for Interpretable Knowledge Tracing and Beyond [55.984684518346924]
  We recast Knowledge Tracing as an inverse problem: learning the minimum natural-language summary that makes past answers explainable and future answers predictable.<n>Our Language Bottleneck Model (LBM) consists of an encoder LLM that writes an interpretable knowledge summary and a frozen decoder LLM that must reconstruct and predict student responses using only that summary text.<n> Experiments on synthetic arithmetic benchmarks and the large-scale Eedi dataset show that LBMs rival the accuracy of state-of-the-art KT and direct LLM methods while requiring orders-of-magnitude fewer student trajectories.
  arXiv  Detail & Related papers  (2025-06-20T13:21:14Z)
• Decomposing MLP Activations into Interpretable Features via Semi-Nonnegative Matrix Factorization [17.101290138120564]
  Current methods rely on dictionary learning with sparse autoencoders (SAEs)<n>Here, we tackle these limitations by directly decomposing activations with semi-nonnegative matrix factorization (SNMF)<n>Experiments on Llama 3.1, Gemma 2 and GPT-2 show that SNMF derived features outperform SAEs and a strong supervised baseline (difference-in-means) on causal steering.
  arXiv  Detail & Related papers  (2025-06-12T17:33:29Z)
• Predictive Prompt Analysis [18.90591503793723]
  Large Language Models (LLMs) are machine learning models that have seen widespread adoption due to their capability of handling previously difficult tasks.<n>We argue it would be useful to perform predictive prompt analysis', in which an automated technique would perform a quick analysis of a prompt.<n>We present Syntactic Prevalence Analyzer (SPA), a predictive prompt analysis approach based on sparse autoencoders (SAEs)
  arXiv  Detail & Related papers  (2025-01-31T04:34:43Z)
• SPaR: Self-Play with Tree-Search Refinement to Improve Instruction-Following in Large Language Models [88.29990536278167]
  We introduce SPaR, a self-play framework integrating tree-search self-refinement to yield valid and comparable preference pairs.<n>Our experiments show that a LLaMA3-8B model, trained over three iterations guided by SPaR, surpasses GPT-4-Turbo on the IFEval benchmark without losing general capabilities.
  arXiv  Detail & Related papers  (2024-12-16T09:47:43Z)
• Improving Dictionary Learning with Gated Sparse Autoencoders [8.3037652157611]
  Gated Sparse Autoencoder (Gated SAE) is a technique for unsupervised discovery of interpretable features in language models' (LMs) activations. In SAEs, the L1 penalty used to encourage sparsity introduces many undesirable biases, such as shrinkage. In training SAEs on LMs of up to 7B parameters, Gated SAEs solve shrinkage, and require half as many firing features to achieve comparable reconstruction fidelity.
  arXiv  Detail & Related papers  (2024-04-24T17:47:22Z)
• In-context Learning Generalizes, But Not Always Robustly: The Case of Syntax [36.98247762224868]
  In-context learning (ICL) is now a common method for teaching large language models (LLMs) new tasks. Do models infer the underlying structure of the task defined by the context, or do they rely on superficial generalizations that only generalize to identically distributed examples? In experiments with models from the GPT, PaLM, and Llama 2 families, we find large variance across LMs. The variance is explained more by the composition of the pre-training corpus and supervision methods than by model size.
  arXiv  Detail & Related papers  (2023-11-13T23:52:43Z)
• Understanding Emergent In-Context Learning from a Kernel Regression Perspective [55.95455089638838]
  Large language models (LLMs) have initiated a paradigm shift in transfer learning.<n>This paper proposes a kernel-regression perspective of understanding LLMs' ICL bahaviors when faced with in-context examples.<n>We find that during ICL, the attention and hidden features in LLMs match the behaviors of a kernel regression.
  arXiv  Detail & Related papers  (2023-05-22T06:45:02Z)
• An Interpretability Evaluation Benchmark for Pre-trained Language Models [37.16893581395874]
  We propose a novel evaluation benchmark providing with both English and Chinese annotated data. It tests LMs abilities in multiple dimensions, i.e., grammar, semantics, knowledge, reasoning and computation. It contains perturbed instances for each original instance, so as to use the rationale consistency under perturbations as the metric for faithfulness.
  arXiv  Detail & Related papers  (2022-07-28T08:28:09Z)
• Fast Few-Shot Classification by Few-Iteration Meta-Learning [173.32497326674775]
  We introduce a fast optimization-based meta-learning method for few-shot classification. Our strategy enables important aspects of the base learner objective to be learned during meta-training. We perform a comprehensive experimental analysis, demonstrating the speed and effectiveness of our approach.
  arXiv  Detail & Related papers  (2020-10-01T15:59:31Z)

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.