Heteroskedastic Signals in Budgeted LLM Verification: Structural Heterogeneity Limits Optimization Gains

• URL: http://arxiv.org/abs/2606.15841v1
• Date: Sun, 14 Jun 2026 14:45:57 GMT
• Title: Heteroskedastic Signals in Budgeted LLM Verification: Structural Heterogeneity Limits Optimization Gains
• Abstract summary: Large language model (LLM) systems increasingly use uncertainty signals to allocate limited computation across verification, test-time scaling, tool execution, and other selective-compute decisions.<n>We show that uncertainty quality is heteroskedastic across cost strata, with some regions exhibiting near-random discriminability despite concentrating many errors.
• Score: 4.560961083890857
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language model (LLM) systems increasingly use uncertainty signals to allocate limited computation across verification, test-time scaling, tool execution, and other selective-compute decisions. Such policies rely on a \emph{global signal comparability assumption}: equal scores should carry comparable decision value across inputs. Using budgeted verification as a controlled diagnostic setting, we identify a failure mode of this assumption: uncertainty quality is heteroskedastic across cost strata, with some regions exhibiting near-random discriminability despite concentrating many errors. Under an explicit local model, we characterize the resulting distortion of global allocation and show that its upper bound scales with cross-stratum signal-quality dispersion. We separate weak signals, optimization instability, and structural heterogeneity through a controlled intervention hierarchy: Threshold, MP-Adapt, MP-Strat, and a deliberately simple cost-stratified thresholding intervention (CST). Across MBPP and MATH using Qwen3-8B, LLaMA3-8B, and GPT-4o-mini, global online adaptation yields inconsistent gains over static thresholding; MP-Strat partially recovers performance, while CST improves hit rate by up to 17 percentage points in strongly heterogeneous settings without gradient updates. These results identify structural heterogeneity, rather than optimizer weakness alone, as the primary bottleneck in the observed settings. More broadly, misaligned feedback structure cannot always be repaired by stronger optimization.

Related papers

• SPARC: Single-Pass Scaling for Motion Forecasting with Conformal Bayesian Last Layers [1.4452289368758382]
  Single-Pass Adaptive Risk is a structured, calibrated, and efficient uncertainty estimate for human motion forecasters.<n>A deterministic backbone predicts the future mean, and a conjugate Bayesian last layer converts time-domain feature leverage into an analytic horizon-wise scale.<n>It ranks first on NLL and on the combined MPJPE+NLL criterion while retaining competitive point accuracy and efficient calibrated tubes.
  arXiv  Detail & Related papers  (2026-08-21T07:20:22Z)
• Distribution-Aware Robust Bilevel Optimization: Quantile-Guided Huber Updates in Two-Timescale Stochastic Approximation [16.106678611834102]
  Bilevel optimization (BLO) is fundamental to hierarchical decision-making but suffers from critical instability under heavy-tailed noise.<n>Existing techniques typically rely on magnitude checks, which fail to distinguish informative geometric signals from outliers.
  arXiv  Detail & Related papers  (2026-06-21T10:57:37Z)
• Clipping Bottleneck: Stabilizing RLVR via Stochastic Recovery of Near-Boundary Signals [83.0127582612634]
  Near-boundary Rescue (NSR) is a minimal, plug-and-play modification that retains out-of-bound tokens to recover lost signals.<n>NSR substantially improves training stability and delivers consistent gains over strong baselines such as DAPO and GSPO.
  arXiv  Detail & Related papers  (2026-05-21T16:45:31Z)
• UnGAP: Uncertainty-Guided Affine Prompting for Real-Time Crack Segmentation [16.902729686979303]
  Uncertainty-Prompted Feature Modulator (UPFM) treats aleatoric uncertainty as an active visual prompt rather than a mere output.<n>UnGAP balances superior segmentation accuracy with real-time inference speed.
  arXiv  Detail & Related papers  (2026-05-04T09:18:57Z)
• Too Correct to Learn: Reinforcement Learning on Saturated Reasoning Data [55.84428098924793]
  We propose Constrained Uniform Top-K Sampling (CUTS), a parameter-free decoding strategy enforcing structure-preserving exploration.<n>We integrate this into Mixed-CUTS, a training framework synergizing exploitative and exploratory rollouts to amplify intra-group advantage variance.<n> Notably, Mixed-CUTS improves Pass@1 accuracy on the challenging AIME25 benchmark by up to 15.1% over standard GRPO.
  arXiv  Detail & Related papers  (2026-04-20T16:43:28Z)
• Beyond Semantic Priors: Mitigating Optimization Collapse for Generalizable Visual Forensics [12.973951916863989]
  We identify a failure mode termed Optimization Collapse, where detectors degenerate to random guessing on non-semantic forgeries.<n>We propose a Contrastive Regional Injection Transformer (CoRIT) which integrates a computationally efficient Contrastive Gradient Proxy (CGP) with three training-free strategies.<n>CoRIT mitigates optimization collapse and achieves state-of-the-art generalization across cross-domain and universal forgery benchmarks.
  arXiv  Detail & Related papers  (2026-03-25T08:03:30Z)
• ODAR: Principled Adaptive Routing for LLM Reasoning via Active Inference [60.958331943869126]
  ODAR-Expert is an adaptive routing framework that optimize the accuracy-efficiency trade-off via principled resource allocation.<n>We show strong and consistent gains, including 98.2% accuracy on MATH and 54.8% on Humanity's Last Exam.
  arXiv  Detail & Related papers  (2026-02-27T05:22:01Z)
• Representation Geometry as a Diagnostic for Out-of-Distribution Robustness [0.5978543974412219]
  We propose a geometry-based diagnostic framework that constructs class-conditional mutual k-nearest-neighbor graphs from in-distribution embeddings.<n>We find that lower spectral complexity and higher mean curvature consistently predict stronger out-of-distribution (OOD) accuracy across checkpoints.<n>Our results demonstrate that representation geometry enables interpretable, label-free diagnosis and supports reliable unsupervised checkpoint selection.
  arXiv  Detail & Related papers  (2026-02-03T19:13:36Z)
• ETR: Outcome-Guided Elastic Trust Regions for Policy Optimization [6.716883192613149]
  We propose textbfElastic textbfTrust textbfETR, a dynamic mechanism that aligns optimization constraints with signal quality.<n>ETR consistently outperforms GRPO, achieving superior accuracy while effectively mitigating policy entropy degradation.
  arXiv  Detail & Related papers  (2026-01-07T09:19:53Z)
• 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)
• Disentangled Federated Learning for Tackling Attributes Skew via Invariant Aggregation and Diversity Transferring [104.19414150171472]
  Attributes skews the current federated learning (FL) frameworks from consistent optimization directions among the clients. We propose disentangled federated learning (DFL) to disentangle the domain-specific and cross-invariant attributes into two complementary branches. Experiments verify that DFL facilitates FL with higher performance, better interpretability, and faster convergence rate, compared with SOTA FL methods.
  arXiv  Detail & Related papers  (2022-06-14T13:12:12Z)
• Amortized Conditional Normalized Maximum Likelihood: Reliable Out of Distribution Uncertainty Estimation [99.92568326314667]
  We propose the amortized conditional normalized maximum likelihood (ACNML) method as a scalable general-purpose approach for uncertainty estimation. Our algorithm builds on the conditional normalized maximum likelihood (CNML) coding scheme, which has minimax optimal properties according to the minimum description length principle. We demonstrate that ACNML compares favorably to a number of prior techniques for uncertainty estimation in terms of calibration on out-of-distribution inputs.
  arXiv  Detail & Related papers  (2020-11-05T08:04:34Z)

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.