Pairwise matrices for sparse autoencoders: single-feature inspection mislabels causal axes

• URL: http://arxiv.org/abs/2605.03160v1
• Date: Mon, 04 May 2026 21:11:21 GMT
• Title: Pairwise matrices for sparse autoencoders: single-feature inspection mislabels causal axes
• Abstract summary: Standard sparse-autoencoder protocol labels each feature from its top-activating contexts and validates by single-feature steering.<n>We propose the pairwise matrix protocol, co-varying steering coefficient with joint condition, and report three findings the standard one-corner protocol misses on Qwen3-1.7B-Instruct, replicated on Gemma-2-2B-it.<n>All three findings reproduce on Gemma with model-specific damage signatures; the matched-geometry control is CI-separated by 10x.
• Score: 2.741152471987327
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The standard sparse-autoencoder (SAE) interpretability protocol labels each feature from its top-activating contexts and validates by single-feature steering. We propose the pairwise matrix protocol, co-varying steering coefficient with joint condition, and report three findings the standard one-corner protocol misses on Qwen3-1.7B-Instruct, replicated on Gemma-2-2B-it. First, a feature labelled "AI self-disclaimer" from its top contexts produces an inverted U-shape under a coefficient sweep: at c=+500 the model substitutes a fluent contemplative-philosopher voice for the disclaimer. Two further features anchor the criterion (one monotonic, one pure breakdown). Second, three near-orthogonal cluster-specific features that individually steer a philosophy-of-mind register, jointly suppressed at c=-500, damage grounded composition on recipes and engine explanations as well as introspective prompts; single-feature suppression at the same magnitude leaves controls intact. Third, a matched-geometry comparison of single-feature, joint, and random-direction perturbations (norm ~1.55, cosine ~0.64) yields three distinct output regimes: single-feature substitutes strategy filler, random direction substitutes diverse content, joint suppression alone produces placeholder text. Coherence loss is direction-pattern-dependent, not magnitude-dependent. All three findings reproduce on Gemma with model-specific damage signatures; the matched-geometry control is CI-separated by ~10x. The pipeline also locates a top causally responsible feature in Llama-3.1-8B-Instruct.

Related papers

• SeGDeP: Semantic- and Geometric-Aware Decoupled Prompts for Reasoning Segmentation [51.46828526392219]
  Reasoning segmentation converts an implicit linguistic conclusion into a precise mask.<n>Existing MLLM-segmenter interfaces either use a special trigger or compress both signals into one context.<n>We present SeGDeP, an explicit what-where interface.
  arXiv  Detail & Related papers  (2026-09-08T15:11:32Z)
• Feature Priming in Online Linear Regression: Sparse-Regret Lower Bounds and a Tight Univariate Rate [47.36630149538994]
  In high-dimensional online prediction, the best predictor may depend on only a few features, so regret should scale with sparsity rather than the ambient dimension.<n>Warmuth and Amid asked at COLT 2023 whether any of three such rules admits a competitive online regret guarantee.<n>Our analysis identifies a common obstruction: cheap nuisance causes the refit to underweight the truly predictive coordinate.
  arXiv  Detail & Related papers  (2026-08-18T09:32:03Z)
• Learning Auditable Classifier Models: Source-Disjoint Tree Ensembles [0.0]
  Predictive models in clinical and regulated settings must be accurate and fully auditable.<n>We introduce Residual Pattern Tree Ensemble (RPTE), a three-stage learning approach.<n>RPTE builds on three key principles: bounded feature budget, source disjointness, and separate coefficient estimation.
  arXiv  Detail & Related papers  (2026-08-16T12:56:49Z)
• Structuring Semantic Embeddings for Principle Evaluation: A Prototype-Guided Contrastive Learning Approach [58.15427952222424]
  General-purpose text embeddings are widely deployed for such tasks, but broad semantic similarity can place semantically similar yet task-distinct examples.<n>We introduce Prototype-Guided Contrastive Learning (PGCL), a prototype-guided geometric regularization module built on top of frozen text embeddings.
  arXiv  Detail & Related papers  (2026-08-15T13:19:25Z)
• From Geometric Recovery to Causal Validation: A Reproducible Audit of Sparse Autoencoder Features, from Superposition Geometry to Causal Inertness [0.0]
  Sparse autoencoders (SAEs) are the standard for decomposing superposed neural representations into interpretable features.<n>We show this conflates two distinct claims: decoder-geometry alignment and encoder-activation behavior.<n>We package the method as sae-causal-audit, a model-agnostic instrument with a deterministic pipeline.
  arXiv  Detail & Related papers  (2026-07-13T21:18:52Z)
• Speculative Refinement: A Hybrid Autoregressive Diffusion Decoding Strategy and Its Behavior Across Benchmarks [1.9800951131982487]
  We use Speculative Refinement (SpecRef) to warm-start a masked diffusion language model from an AR draft.<n>We evaluate SpecRef across six benchmarks (HumanEval, MBPP, GSM8K, BBH, ARC-Challenge, HellaSwag) with three distinct evaluation protocols.<n>These observations apply to any multi-stage or non-autoregressive generation pipeline and point toward more diagnostic evaluation practices.
  arXiv  Detail & Related papers  (2026-06-25T18:52:24Z)
• Verified Detection and Prevention of Concurrency Anomalies in Multi-Agent Large Language Model Systems [0.0]
  Multi-agent LLM systems state through memory stores, vector indices, and tool registries.<n>We model such sharing as long-running read-generate-write operations under deterministic-generation semantics.<n>We formalize four anomalies in TLA+: stale-generation, phantom-tool, causal-cascade, and tool-effect reordering.
  arXiv  Detail & Related papers  (2026-06-15T18:19:34Z)
• Hierarchical Consistency Learning for Test-time Adaptation in Camouflage Perception [50.278200968044665]
  Camouflaged object detection (COD) aims to localize targets that exhibit minimal perceptual differences from backgrounds through physical attributes.<n>Existing methods, constrained by the static train-then-freeze paradigm, suffer from domain rigidity and dependency annotation.<n>We propose the hierarchical consistency learning framework, which integrates test-time adaptation for dynamic representation recalibration.
  arXiv  Detail & Related papers  (2026-05-25T09:57:46Z)
• Every Component is a Lookup: Token Attribution and Composition from a Single Decomposition [1.3290717106567922]
  We exploit a shared key-value template $(S)U$ to develop Unpack, a backward recursion that decomposes credit through both sublayers.<n>On GPT-2 small, the method recovers all three composition connections described by Wang et al. (2023), including the mode-specific routing of each connection.<n>Across the Pythia family from 160M to 6.9B parameters, this suppression pattern is consistently recovered at every scale.
  arXiv  Detail & Related papers  (2026-05-22T09:03:01Z)
• Collapse-Free Prototype Readout Layer for Transformer Encoders [0.0]
  DDCL-Attention is a prototype-based readout layer for transformer encoders.<n>It produces compact token summaries at linear complexity in sequence length.<n>Same framework supports three uses: a final readout layer, a differentiable codebook extending VQ-VAE, and a hierarchical document compressor.
  arXiv  Detail & Related papers  (2026-04-04T20:23:21Z)
• Deterministic Fuzzy Triage for Legal Compliance Classification and Evidence Retrieval [0.0]
  Legal teams increasingly use machine learning to triage large volumes of contractual evidence.<n>Many models are opaque, non-deterministic, and difficult to align with frameworks such as HIPAA or NERC-CIP.<n>We study a simple, reproducible alternative based on deterministic dual encoders and transparent fuzzy triage bands.
  arXiv  Detail & Related papers  (2026-03-08T00:31:34Z)
• From Feature Interaction to Feature Generation: A Generative Paradigm of CTR Prediction Models [81.43473418572567]
  Click-Through Rate (CTR) prediction is a core task in recommendation systems.<n>We propose a novel generative framework to address embedding dimensional collapse and information redundancy.<n>We show that SFG consistently mitigates embedding collapse and reduces information redundancy, while yielding substantial performance gains.
  arXiv  Detail & Related papers  (2025-12-16T03:17:18Z)
• Ensemble Threshold Calibration for Stable Sensitivity Control [0.0]
  We present an end-to-end framework that achieves exact recall with sub-percent variance over tens of millions of geometry pairs.<n>Our approach consistently hits a recall target within a small error, decreases redundant verifications relative to other calibrations, and runs end-to-end on a single TPU v3 core.
  arXiv  Detail & Related papers  (2025-10-02T15:22:28Z)
• Unsupervised Conformal Inference: Bootstrapping and Alignment to Control LLM Uncertainty [49.19257648205146]
  We propose an unsupervised conformal inference framework for generation.<n>Our gates achieve close-to-nominal coverage and provide tighter, more stable thresholds than split UCP.<n>The result is a label-free, API-compatible gate for test-time filtering.
  arXiv  Detail & Related papers  (2025-09-26T23:40:47Z)
• Label Distributionally Robust Losses for Multi-class Classification: Consistency, Robustness and Adaptivity [55.29408396918968]
  We study a family of loss functions named label-distributionally robust (LDR) losses for multi-class classification. Our contributions include both consistency and robustness by establishing top-$k$ consistency of LDR losses for multi-class classification. We propose a new adaptive LDR loss that automatically adapts the individualized temperature parameter to the noise degree of class label of each instance.
  arXiv  Detail & Related papers  (2021-12-30T00:27:30Z)
• S3: Supervised Self-supervised Learning under Label Noise [53.02249460567745]
  In this paper we address the problem of classification in the presence of label noise. In the heart of our method is a sample selection mechanism that relies on the consistency between the annotated label of a sample and the distribution of the labels in its neighborhood in the feature space. Our method significantly surpasses previous methods on both CIFARCIFAR100 with artificial noise and real-world noisy datasets such as WebVision and ANIMAL-10N.
  arXiv  Detail & Related papers  (2021-11-22T15:49:20Z)
• A Quadruplet Loss for Enforcing Semantically Coherent Embeddings in Multi-output Classification Problems [5.972927416266617]
  This paper describes one objective function for learning semantically coherent feature embeddings in multi-output classification problems. We consider the problems of identity retrieval and soft biometrics labelling in visual surveillance environments.
  arXiv  Detail & Related papers  (2020-02-26T17:18:53Z)

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.