The Matching Principle: A Geometric Theory of Loss Functions for Nuisance-Robust Representation Learning

• URL: http://arxiv.org/abs/2605.22800v2
• Date: Sat, 23 May 2026 08:58:51 GMT
• Title: The Matching Principle: A Geometric Theory of Loss Functions for Nuisance-Robust Representation Learning
• Abstract summary: We prove optimality in a linear-Gaussian model (Thm. A), necessity of range coverage for any quadratic penalty that zeros deployment drift (Thm. G)<n>Thirteen blocks (ML through Qwen2.5-7B) test matched vs isotropic vs wrong-direction penalties on geometry and deployment drift.<n>We claim a falsifiable design recipe: estimate Sigma_task, match Sigma', run the controls, report task and geometry separately.
• Score: 0.5854803320592717
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Robustness, domain adaptation, photometric/occlusion invariance, sensor drift, and alignment style are treated as separate literatures with separate method families. Under label-preserving deployment shift they share one geometric object: the covariance Sigma_task = Cov_{Q_n}(n) of ways inputs can change without changing the label. CORAL, adversarial training, augmentation, metric learning, Jacobian penalties, and alignment constraints are not independent tricks--they are estimators of Sigma_task. Fix that object and the Jacobian penalty is pinned by a matrix Sigma' whose range must cover range(Sigma_task)--the matching principle. We prove optimality in a linear-Gaussian model (Thm. A), necessity of range coverage for any quadratic penalty that zeros deployment drift (Thm. G), and the same dichotomy at global minima (Thm. A*_global). Wrong-direction/signal-aligned controls (Lemma C; Cor. E/E*) and seven estimators (Lemmas D1--D7), plus label-free TDI, yield a falsifiable recipe when Sigma_task must be learned. Thirteen blocks (ML through Qwen2.5-7B) test matched vs isotropic vs wrong-direction penalties on geometry and deployment drift. Twelve match theory where identifiability holds; Office-31 is a named eigengap failure. Partial passes: geometry can improve without every headline task metric moving. A pilot 7B DPO run (one epoch, 240 pairs): matched style-PMH preserves Style TDI where standard DPO degrades it. We do not claim standard training reaches global minima (assumption (O) is open), that estimated Sigma_task is always identifiable, or dominance on every leaderboard. We claim a falsifiable design recipe: estimate Sigma_task, match Sigma', run the controls, report task and geometry separately.

Related papers

• The Filter Metric is Safety-Critical: Phantom Advantages in Group-Relative RL under Shaped Rewards [1.667501390377158]
  Group-relative policy optimization can discard nocontrast rollout groups through dynamic sampling.<n>We identify and quantify a metricpredicate mismatch under composite shaped rewards.
  arXiv  Detail & Related papers  (2026-09-12T10:36:55Z)
• The Cost of Adaptivity: Matching Lower Bounds Across Learning Problems [1.0742675209112622]
  We formalize nuisance adaptation via a slice-normalized minimax ratio.<n>We also define the cost of expanding from one preannounced Gaussian query to arbitrary post-hoc inspection.
  arXiv  Detail & Related papers  (2026-08-09T17:19:11Z)
• When Model Merging Rivals Joint Multi-Task Reinforcement Learning: A Task-Vector Geometry Analysis [0.16921396880325779]
  Training difficulty-1 and difficulty-2 Qwen3-8B specialists on the AppWorld agent benchmark with LOOP.<n>We merge them (TIES, RAM+) and pit the result against a jointly trained model on the same data.<n>On task-goal completion, merging matches joint RL -- and every merge variant is statistically indistinguishable.
  arXiv  Detail & Related papers  (2026-07-17T15:41:09Z)
• Bayesian Wind Tunnels for Model Selection [0.4779196219827507]
  We show that transformers can perform exact Bayesian filtering within a fixed hypothesis class.<n>We then identify a sharp perceptual access condition: when the discnative statistic requires arithmetic, model selection succeeds with integer tokens.<n>A stationarity control confirms the operative factor: stable semantics, not integer identity, enable circuit compilation.
  arXiv  Detail & Related papers  (2026-07-01T04:05:08Z)
• Matrix-Decoupled Concentration for Autoregressive Sequences: Dimension-Free Guarantees for Sparse Long-Context Rewards [0.36260136172126667]
  We show how to establish tight concentration bounds for Sequence-level evaluations in autoregressive Large Language Models.<n>By preserving the coordinate-wise sparsity of rewards within a strictly causal framework, we provide a rigorous mathematical justification for the stability of long-context reasoning.
  arXiv  Detail & Related papers  (2026-05-07T11:12:59Z)
• Supervised Learning Has a Necessary Geometric Blind Spot: Theory, Consequences, and Minimal Repair [0.5854803320592717]
  We show that any encoder minimizing supervised loss must retain non-zero sensitivity along directions correlated with training labels.<n>We call this the geometric blind spot of supervised learning.
  arXiv  Detail & Related papers  (2026-04-23T08:03:33Z)
• 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)
• Label-Free Cross-Task LoRA Merging with Null-Space Compression [50.63908869296697]
  We introduce Null-Space Compression (NSC) Merging, a label-free, output-agnostic method that sets merge weights from adapter geometry.<n>NSC achieves state-of-the-art performance across twenty heterogeneous vision tasks with balanced gains where prior methods overfit subsets of tasks.
  arXiv  Detail & Related papers  (2026-03-27T11:34:41Z)
• Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learning [17.384089089363382]
  We identify a root cause that existing methods overlook: the uniform penalization of errors.<n>Current approaches treat all incorrect rollouts within a group identically.<n>We propose the Asymmetric Confidence-aware Error Penalty (ACE)
  arXiv  Detail & Related papers  (2026-02-24T22:46:43Z)
• Stability and Generalization of Push-Sum Based Decentralized Optimization over Directed Graphs [55.77845440440496]
  Push-based decentralized communication enables optimization over communication networks, where information exchange may be asymmetric.<n>We develop a unified uniform-stability framework for the Gradient Push (SGP) algorithm.<n>A key technical ingredient is an imbalance-aware generalization bound through two quantities.
  arXiv  Detail & Related papers  (2026-02-24T05:32:03Z)
• Optimal Unconstrained Self-Distillation in Ridge Regression: Strict Improvements, Precise Asymptotics, and One-Shot Tuning [61.07540493350384]
  Self-distillation (SD) is the process of retraining a student on a mixture of ground-truth and the teacher's own predictions.<n>We show that for any prediction risk, the optimally mixed student improves upon the ridge teacher for every regularization level.<n>We propose a consistent one-shot tuning method to estimate $star$ without grid search, sample splitting, or refitting.
  arXiv  Detail & Related papers  (2026-02-19T17:21:15Z)
• The Emergence of Lab-Driven Alignment Signatures: A Psychometric Framework for Auditing Latent Bias and Compounding Risk in Generative AI [0.0]
  This paper introduces a novel auditing framework to quantify latent trait estimation under ordinal uncertainty.<n>The research audits nine leading models across dimensions including Optimization Bias, Sycophancy, and Status-Quo Legitimization.
  arXiv  Detail & Related papers  (2026-02-19T06:56:01Z)
• The Procrustean Bed of Time Series: The Optimization Bias of Point-wise Loss [53.542743390809356]
  This paper aims to provide a first-principles analysis of the Expectation of Optimization Bias (EOB)<n>Our analysis reveals a fundamental paradigm paradox: the more deterministic and structured the time series, the more severe the bias by point-wise loss function.<n>We present a concrete solution that simultaneously achieves both principles via DFT or DWT.
  arXiv  Detail & Related papers  (2025-12-21T06:08:22Z)
• Ordinal Label-Distribution Learning with Constrained Asymmetric Priors for Imbalanced Retinal Grading [9.147336466586017]
  Diabetic retinopathy grading is inherently ordinal and long-tailed.<n>We propose the Constrained Asymmetric Prior Wasserstein Autoencoder (CAP-WAE)<n>CAP-WAE consistently achieves state-of-the-art Quadratic Weighted Kappa, accuracy, and macro-F1.
  arXiv  Detail & Related papers  (2025-09-30T11:58:49Z)
• Towards Self-Supervised Covariance Estimation in Deep Heteroscedastic Regression [102.24287051757469]
  We study self-supervised covariance estimation in deep heteroscedastic regression.<n>We derive an upper bound on the 2-Wasserstein distance between normal distributions.<n>Experiments over a wide range of synthetic and real datasets demonstrate that the proposed 2-Wasserstein bound coupled with pseudo label annotations results in a computationally cheaper yet accurate deep heteroscedastic regression.
  arXiv  Detail & Related papers  (2025-02-14T22:37:11Z)
• Generating Unbiased Pseudo-labels via a Theoretically Guaranteed Chebyshev Constraint to Unify Semi-supervised Classification and Regression [57.17120203327993]
  threshold-to-pseudo label process (T2L) in classification uses confidence to determine the quality of label. In nature, regression also requires unbiased methods to generate high-quality labels. We propose a theoretically guaranteed constraint for generating unbiased labels based on Chebyshev's inequality.
  arXiv  Detail & Related papers  (2023-11-03T08:39:35Z)
• Benign Overfitting of Constant-Stepsize SGD for Linear Regression [122.70478935214128]
  inductive biases are central in preventing overfitting empirically. This work considers this issue in arguably the most basic setting: constant-stepsize SGD for linear regression. We reflect on a number of notable differences between the algorithmic regularization afforded by (unregularized) SGD in comparison to ordinary least squares.
  arXiv  Detail & Related papers  (2021-03-23T17:15:53Z)
• Joint Noise-Tolerant Learning and Meta Camera Shift Adaptation for Unsupervised Person Re-Identification [60.36551512902312]
  unsupervised person re-identification (re-ID) aims to learn discriminative models with unlabeled data. One popular method is to obtain pseudo-label by clustering and use them to optimize the model. In this paper, we propose a unified framework to solve both problems.
  arXiv  Detail & Related papers  (2021-03-08T09:13:06Z)
• A Unified Joint Maximum Mean Discrepancy for Domain Adaptation [73.44809425486767]
  This paper theoretically derives a unified form of JMMD that is easy to optimize. From the revealed unified JMMD, we illustrate that JMMD degrades the feature-label dependence that benefits to classification. We propose a novel MMD matrix to promote the dependence, and devise a novel label kernel that is robust to label distribution shift.
  arXiv  Detail & Related papers  (2021-01-25T09:46:14Z)
• Coping with Label Shift via Distributionally Robust Optimisation [72.80971421083937]
  We propose a model that minimises an objective based on distributionally robust optimisation (DRO) We then design and analyse a gradient descent-proximal mirror ascent algorithm tailored for large-scale problems to optimise the proposed objective.
  arXiv  Detail & Related papers  (2020-10-23T08:33:04Z)
• Unifying supervised learning and VAEs -- coverage, systematics and goodness-of-fit in normalizing-flow based neural network models for astro-particle reconstructions [0.0]
  Statistical uncertainties, coverage, systematic uncertainties or a goodness-of-fit measure are often not calculated. We show that a KL-divergence objective of the joint distribution of data and labels allows to unify supervised learning and variational autoencoders. We discuss how to calculate coverage probabilities without numerical integration for specific "base-ordered" contours.
  arXiv  Detail & Related papers  (2020-08-13T11:28:57Z)

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.