Reference-Based Bias Detection in LLMs via Relative Representations of Hidden States

• URL: http://arxiv.org/abs/2609.10060v1
• Date: Wed, 09 Sep 2026 11:33:35 GMT
• Title: Reference-Based Bias Detection in LLMs via Relative Representations of Hidden States
• Abstract summary: We propose a reference-based method that audits bias in hidden-state representations across related model variants.<n>Because fine-tuning reshapes representation geometry, absolute hidden states are not directly comparable.<n>We measure how target groups shift in their association with positive and negative attributes.
• Score: 4.505185728893222
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Existing bias auditing methods typically rely on model outputs, requiring costly benchmarks or judge models and potentially missing internal shifts that never appear in generated text. We propose a reference-based method that audits bias in hidden-state representations across related model variants, for example before and after fine-tuning. Because fine-tuning reshapes representation geometry, absolute hidden states are not directly comparable, so we encode each sentence by its similarities to a fixed set of anchor sentences, yielding relative representations in a shared comparison space. There we measure how target groups shift in their association with positive and negative attributes, a quantity we call the Representational Bias Shift $ΔB$. Across three model families and the WildGuardMix, DecodingTrust and ToxiGen benchmarks, $ΔB$ correlates with output-level bias change in 15 of the 18 settings we test, reaching $|r| = 0.84$ ($p < 0.001$) under full fine-tuning and becoming more model-dependent under parameter-efficient adaptation. Thresholding $ΔB$ detects checkpoints whose bias increased with ROC AUC between $0.65$ and $0.99$, and on WildGuardMix and DecodingTrust it separates them better than a SEAT-based baseline for all three families. $ΔB$ is also stable under changes to the anchor set, attribute sets and target templates. Our method requires no task-specific evaluation data and audits a model in about three minutes, using $3$-$50\times$ less compute than the output-level benchmarks considered here. We view it as complementary to output-based auditing rather than a replacement for it.

Related papers

• Which Model Is Actually Serving You? IRIS: Budgeted Black-Box Auditing of Model Substitution and Routing Dilution in LLM Gateways [6.964275075635626]
  We present $mathrmIRIS$, an audit that needs only the returned text.<n>A cheap pilot fits the exponential query-error decay and freezes that budget before any suspect query is issued.<n>$mathrmIRIS$ matches or beats detection on shared tasks, and adaptive allocation lifts the matched-budget target-hit rate from $73$% to $87$%.
  arXiv  Detail & Related papers  (2026-07-23T02:32:08Z)
• Statistically Indistinguishable, Operationally Distinct: A Formal Barrier for Tabular Foundation Models [6.240573495762223]
  We show that Tabular foundation models cannot reason about data produced by running systems without access to the rules that govern them.<n>In three matched $100$-state frontier large-lemma-model (LLM) runs, models given the schema, trigger source, rule tables, and state files classify at most $2/50$ legal states as Legal.<n>The barrier is identifiability, not capacity scale, data, and richer features cannot cross it without operational grounding.
  arXiv  Detail & Related papers  (2026-06-27T21:14:06Z)
• Scaling Laws for Agent Harnesses via Effective Feedback Compute [53.68149869349268]
  emphEffective Feedback Compute (EFC) is a trace-level scaling coordinate that credits feedback only when it is informative, valid, non-redundant, and retained for subsequent decisions.<n>EFC-based coordinates consistently predict failure rates better than raw-compute baselines.
  arXiv  Detail & Related papers  (2026-05-28T09:45:47Z)
• ChainzRule: Sample-Efficient, Robust Deep Learning Across Tabular, NLP, and Vision Tasks [0.0]
  Production deep learning systems across enterprise domains operate under constraints that academic benchmarks routinely obscure.<n>We present ChainzRule (CR), a neural architecture replacing typical activations with learnable layers governed by Differential Regularization (DREG)<n>CR achieves $85.71% pm 2.01%$ on Pima Diabetes, $46.20% pm 0.37%$ on SST-5 sentiment classification with a frozen encoder, and $70.17%$ on Full ordinal regression with 3.2M parameters.
  arXiv  Detail & Related papers  (2026-05-23T01:52:50Z)
• Representation Without Reward: A JEPA Audit for LLM Fine-Tuning [1.2691047660244335]
  Joint-embedding predictive architectures (JEPAs) propose that a model should learn more useful abstractions when trained to predict latent representations rather than observed outputs.<n>For autoregressive language-model fine-tuning the principle entails a stricter requirement: the induced hidden-state geometry must reach the language-model head emphand improve the decoded task metric.<n>We test that requirement under a fixed Llama-3.2-1B-Instruct LoRA harness on natural-language-to-regex generation.
  arXiv  Detail & Related papers  (2026-05-14T20:27:32Z)
• The Surprising Universality of LLM Outputs: A Real-Time Verification Primitive [0.0]
  CPU-only scoring primitive runs at 2.6 microseconds per token.<n> token rank-frequency distributions converge to the same two- parameter Mandelbrot ranking distribution.<n>We derive a single-pass scoring primitive that composes with model log probabilities when available and degrades to a rank-only mode usable on closed APIs.
  arXiv  Detail & Related papers  (2026-04-28T13:35:31Z)
• $V_1$: Unifying Generation and Self-Verification for Parallel Reasoners [69.66089681814013]
  $V_$ is a framework that unifies generation and verification through efficient pairwise ranking.<n>$V_$-Infer improves Pass@1 by up to $10%$ over pointwise verification.<n>$V_$-PairRL achieves $7$--$9%$ test-time scaling gains over standard RL and pointwise joint training.
  arXiv  Detail & Related papers  (2026-03-04T17:22:16Z)
• Learning Accurate Segmentation Purely from Self-Supervision [87.78965637247107]
  Selfment is a fully self-supervised framework that segments objects directly from raw images without human labels.<n>Selfment sets new state-of-the-art (SoTA) results across multiple benchmarks.
  arXiv  Detail & Related papers  (2026-02-27T07:36:32Z)
• From Global to Granular: Revealing IQA Model Performance via Correlation Surface [83.65597122328133]
  We present textbfGranularity-Modulated Correlation (GMC), which provides a structured, fine-grained analysis of IQA performance.<n>GMC includes a textbfDistribution Regulator that regularizes correlations to mitigate biases from non-uniform quality distributions.<n>Experiments on standard benchmarks show that GMC reveals performance characteristics invisible to scalar metrics, offering a more informative and reliable paradigm for analyzing, comparing, and deploying IQA models.
  arXiv  Detail & Related papers  (2026-01-29T13:55:26Z)
• Correcting Mean Bias in Text Embeddings: A Refined Renormalization with Training-Free Improvements on MMTEB [5.315831681082424]
  We propose a plug-and-play, training-free and lightweight solution called Renormalization.<n>We show that renormalization consistently and statistically significantly improves the performance of existing models.
  arXiv  Detail & Related papers  (2025-11-14T07:51:59Z)
• Reinforcement Learning with Verifiable yet Noisy Rewards under Imperfect Verifiers [90.50039419576807]
  Reinforcement Learning with Verifiable Rewards (RLVR) trains policies against automated verifiers to avoid costly human labeling.<n>To reduce vulnerability to verifier hacking, many RLVR systems collapse rewards to binary $0,1$ during training.<n>This choice carries a cost: it introduces textitfalse negatives (rejecting correct answers, FNs) and textitfalse positives (accepting incorrect ones, FPs)
  arXiv  Detail & Related papers  (2025-10-01T13:56:44Z)
• Beyond Invariance: Test-Time Label-Shift Adaptation for Distributions with "Spurious" Correlations [44.99833362998488]
  Changes in the data distribution at test time can have deleterious effects on the performance of predictive models. We propose a test-time label shift correction that adapts to changes in the joint distribution $p(y, z)$ using EM applied to unlabeled samples.
  arXiv  Detail & Related papers  (2022-11-28T18:52:33Z)
• Evaluating Models' Local Decision Boundaries via Contrast Sets [119.38387782979474]
  We propose a new annotation paradigm for NLP that helps to close systematic gaps in the test data. We demonstrate the efficacy of contrast sets by creating them for 10 diverse NLP datasets. Although our contrast sets are not explicitly adversarial, model performance is significantly lower on them than on the original test sets.
  arXiv  Detail & Related papers  (2020-04-06T14:47:18Z)

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.