Breaking the Correlation Plateau: On the Optimization and Capacity Limits of Attention-Based Regressors

• URL: http://arxiv.org/abs/2602.17898v1
• Date: Thu, 19 Feb 2026 23:33:04 GMT
• Title: Breaking the Correlation Plateau: On the Optimization and Capacity Limits of Attention-Based Regressors
• Authors: Jingquan Yan, Yuwei Miao, Peiran Yu, Junzhou Huang,
• Abstract summary: A common but poorly understood phenomenon during training is the PCC plateau: PCC stops improving early in training, even as MSE continues to decrease.<n>We provide the first rigorous theoretical analysis of this behavior, revealing fundamental limitations in both optimization dynamics and model capacity.<n>Motivated by these insights, we propose the Extrapolative Correlation Attention (ECA), which incorporates novel, theoretically-motivated mechanisms to improve the PCC optimization and extrapolate beyond the convex hull.
• Score: 21.002587086609243
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Attention-based regression models are often trained by jointly optimizing Mean Squared Error (MSE) loss and Pearson correlation coefficient (PCC) loss, emphasizing the magnitude of errors and the order or shape of targets, respectively. A common but poorly understood phenomenon during training is the PCC plateau: PCC stops improving early in training, even as MSE continues to decrease. We provide the first rigorous theoretical analysis of this behavior, revealing fundamental limitations in both optimization dynamics and model capacity. First, in regard to the flattened PCC curve, we uncover a critical conflict where lowering MSE (magnitude matching) can paradoxically suppress the PCC gradient (shape matching). This issue is exacerbated by the softmax attention mechanism, particularly when the data to be aggregated is highly homogeneous. Second, we identify a limitation in the model capacity: we derived a PCC improvement limit for any convex aggregator (including the softmax attention), showing that the convex hull of the inputs strictly bounds the achievable PCC gain. We demonstrate that data homogeneity intensifies both limitations. Motivated by these insights, we propose the Extrapolative Correlation Attention (ECA), which incorporates novel, theoretically-motivated mechanisms to improve the PCC optimization and extrapolate beyond the convex hull. Across diverse benchmarks, including challenging homogeneous data setting, ECA consistently breaks the PCC plateau, achieving significant improvements in correlation without compromising MSE performance.

Related papers

• 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)
• Difficulty-Estimated Policy Optimization [38.86673795561421]
  We propose Difficulty-Estimated Policy Optimization (DEPO), a novel framework designed to optimize the efficiency and robustness of reasoning alignment.<n>Our approach significantly lowers the computational barrier for training high-performance reasoning models, offering a more sustainable path for reasoning scaling.
  arXiv  Detail & Related papers  (2026-02-06T04:12:23Z)
• Conformal prediction for full and sparse polynomial chaos expansions [0.5735035463793009]
  Polynomial Chaos Expansions (PCEs) are widely recognized for their efficient computational performance in surrogate modeling.<n>We present the integration of two conformal prediction methods, namely the full conformal and the Jackknife+ approaches, into both full and sparse PCEs.
  arXiv  Detail & Related papers  (2026-01-23T10:46:46Z)
• Learning Causality for Longitudinal Data [1.2691047660244335]
  This thesis develops methods for causal inference and causal representation learning in high-dimensional, time-varying data.<n>The first contribution introduces the Causal Dynamic Variational Autoencoder (CDVAE), a model for estimating Individual Treatment Effects (ITEs)<n>The second contribution proposes an efficient framework for long-term counterfactual regression based on RNNs enhanced with Contrastive Predictive Coding ( CPC) and InfoMax.<n>The third contribution advances CRL by addressing how latent causes manifest in observed variables.
  arXiv  Detail & Related papers  (2025-12-04T16:51:49Z)
• Composable Score-based Graph Diffusion Model for Multi-Conditional Molecular Generation [85.58520120011269]
  We propose Composable Score-based Graph Diffusion model (CSGD), which extends score matching to discrete graphs via concrete scores.<n>We show that CSGD achieves state-of-the-art performance with a 15.3% average improvement in controllability over prior methods.<n>Our findings highlight the practical advantages of score-based modeling for discrete graph generation and its capacity for flexible, multi-property molecular design.
  arXiv  Detail & Related papers  (2025-09-11T13:37:56Z)
• On the Convergence of DP-SGD with Adaptive Clipping [56.24689348875711]
  Gradient Descent with gradient clipping is a powerful technique for enabling differentially private optimization.<n>This paper provides the first comprehensive convergence analysis of SGD with quantile clipping (QC-SGD)<n>We show how QC-SGD suffers from a bias problem similar to constant-threshold clipped SGD but can be mitigated through a carefully designed quantile and step size schedule.
  arXiv  Detail & Related papers  (2024-12-27T20:29:47Z)
• Conformal Risk Minimization with Variance Reduction [37.74931189657469]
  Conformal prediction (CP) is a distribution-free framework for achieving probabilistic guarantees on black-box models.<n>Recent research efforts have focused on optimizing CP efficiency during training.<n>We formalize this concept as the problem of conformal risk minimization.
  arXiv  Detail & Related papers  (2024-11-03T21:48:15Z)
• Prior Constraints-based Reward Model Training for Aligning Large Language Models [58.33118716810208]
  This paper proposes a Prior Constraints-based Reward Model (namely PCRM) training method to mitigate this problem. PCRM incorporates prior constraints, specifically, length ratio and cosine similarity between outputs of each comparison pair, during reward model training to regulate optimization magnitude and control score margins. Experimental results demonstrate that PCRM significantly improves alignment performance by effectively constraining reward score scaling.
  arXiv  Detail & Related papers  (2024-04-01T07:49:11Z)
• A PAC-Bayesian Perspective on the Interpolating Information Criterion [54.548058449535155]
  We show how a PAC-Bayes bound is obtained for a general class of models, characterizing factors which influence performance in the interpolating regime. We quantify how the test error for overparameterized models achieving effectively zero training error depends on the quality of the implicit regularization imposed by e.g. the combination of model, parameter-initialization scheme.
  arXiv  Detail & Related papers  (2023-11-13T01:48:08Z)
• InterMPL: Momentum Pseudo-Labeling with Intermediate CTC Loss [43.39035144463951]
  Momentum PL (MPL) trains a connectionist temporal classification ( CTC)-based model on unlabeled data. CTC is well suited for MPL, or PL-based semi-supervised ASR in general, owing to its simple/fast inference algorithm and robustness against generating collapsed labels. We propose to enhance MPL by introducing intermediate loss, inspired by the recent advances in CTC-based modeling.
  arXiv  Detail & Related papers  (2022-11-02T00:18:25Z)
• A novel Deep Learning approach for one-step Conformal Prediction approximation [0.7646713951724009]
  Conformal Prediction (CP) is a versatile solution that guarantees a maximum error rate given minimal constraints. We propose a novel conformal loss function that approximates the traditionally two-step CP approach in a single step.
  arXiv  Detail & Related papers  (2022-07-25T17:46:09Z)
• Autoregressive Score Matching [113.4502004812927]
  We propose autoregressive conditional score models (AR-CSM) where we parameterize the joint distribution in terms of the derivatives of univariable log-conditionals (scores) For AR-CSM models, this divergence between data and model distributions can be computed and optimized efficiently, requiring no expensive sampling or adversarial training. We show with extensive experimental results that it can be applied to density estimation on synthetic data, image generation, image denoising, and training latent variable models with implicit encoders.
  arXiv  Detail & Related papers  (2020-10-24T07:01:24Z)

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.