Continual Calibration: Coverage Can Collapse Before Accuracy in Lifelong LLM Fine-Tuning

• URL: http://arxiv.org/abs/2604.23987v1
• Date: Mon, 27 Apr 2026 03:03:38 GMT
• Title: Continual Calibration: Coverage Can Collapse Before Accuracy in Lifelong LLM Fine-Tuning
• Abstract summary: Uncertainty reliability can degrade earlier and more sharply than top-1 performance.<n>We study this empirically by measuring conformal coverage and calibration error on sequentially fine-tuned models.<n>We propose calibration replay, a lightweight post-hoc procedure that maintains a task-specific buffer and refits a task-specific conformal threshold.
• Score: 6.908972852063454
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Continual learning for large language models is typically evaluated through accuracy retention under sequential fine-tuning. We argue that this perspective is incomplete, because uncertainty reliability can degrade earlier and more sharply than top-1 performance. We study this empirically by measuring conformal coverage and calibration error on sequentially fine-tuned models across three model families and eight task sequences drawn primarily from classification and multiple-choice benchmarks. Across the classification-style settings we study, coverage loss exceeds accuracy loss by a factor of roughly \(3.4\times \pm 0.5\times\) on average across seeds; in the most pronounced case, coverage drops from \(0.92\) to \(0.61\), while accuracy remains within three points of baseline. Standard continual-learning methods that preserve accuracy do not automatically preserve coverage, and naive calibration baselines recover only part of the gap. We propose calibration replay, a lightweight post-hoc procedure that maintains a task-specific held-out buffer and refits a task-specific conformal threshold under the current model after each update. It adds no training-time gradient cost, uses less than one percent of the memory of ordinary experience replay, and typically restores coverage to within two points of nominal at buffer size \(m = 200\). We accompany the empirical study with a drift decomposition, a finite-sample recovery theorem showing exact conformal validity under exchangeability, and a mixture-validity proposition explaining why pooled thresholds do not suffice. Our guarantees are stated for classification-style tasks with task-specific buffers; extensions to open-ended generation are exploratory.

Related papers

• Nonparametric Distribution Regression Re-calibration [3.0204520109309847]
  Minimizing overall prediction error encourages models to prioritize informativeness over calibration.<n>In safety-critical settings, trustworthy uncertainty estimates are often more valuable than narrow intervals.<n>We propose a novel non-parametric re-calibration algorithm based on conditional kernel mean embeddings.
  arXiv  Detail & Related papers  (2026-02-13T11:48:43Z)
• Temper-Then-Tilt: Principled Unlearning for Generative Models through Tempering and Classifier Guidance [51.532841645285835]
  We study machine unlearning in large generative models by framing the task as density ratio estimation to a target distribution.<n>We show it can fail to faithfully unlearn with finite samples when the forget set represents a sharp, concentrated data distribution.<n>We introduce Temper-Then-Tilt Unlearning (T3-Unlearning), which freezes the base model and applies a two-step inference procedure.
  arXiv  Detail & Related papers  (2026-02-10T19:08:40Z)
• Post-Hoc Split-Point Self-Consistency Verification for Efficient, Unified Quantification of Aleatoric and Epistemic Uncertainty in Deep Learning [5.996056764788456]
  Uncertainty quantification (UQ) is vital for trustworthy deep learning, yet existing methods are either computationally intensive or provide only partial, task-specific estimates.<n>We propose a post-hoc single-forward-pass framework that jointly captures aleatoric and epistemic uncertainty without modifying or retraining pretrained models.<n>Our method applies emphSplit-Point Analysis (SPA) to decompose predictive residuals into upper and lower subsets, computing emphMean Absolute Residuals (MARs) on each side.
  arXiv  Detail & Related papers  (2025-09-16T17:16:01Z)
• Rethinking Early Stopping: Refine, Then Calibrate [49.966899634962374]
  We present a novel variational formulation of the calibration-refinement decomposition.<n>We provide theoretical and empirical evidence that calibration and refinement errors are not minimized simultaneously during training.
  arXiv  Detail & Related papers  (2025-01-31T15:03:54Z)
• Calibrated Uncertainty Quantification for Operator Learning via Conformal Prediction [95.75771195913046]
  We propose a risk-controlling quantile neural operator, a distribution-free, finite-sample functional calibration conformal prediction method. We provide a theoretical calibration guarantee on the coverage rate, defined as the expected percentage of points on the function domain. Empirical results on a 2D Darcy flow and a 3D car surface pressure prediction task validate our theoretical results.
  arXiv  Detail & Related papers  (2024-02-02T23:43:28Z)
• Calibration by Distribution Matching: Trainable Kernel Calibration Metrics [56.629245030893685]
  We introduce kernel-based calibration metrics that unify and generalize popular forms of calibration for both classification and regression. These metrics admit differentiable sample estimates, making it easy to incorporate a calibration objective into empirical risk minimization. We provide intuitive mechanisms to tailor calibration metrics to a decision task, and enforce accurate loss estimation and no regret decisions.
  arXiv  Detail & Related papers  (2023-10-31T06:19:40Z)
• Deep Imbalanced Regression via Hierarchical Classification Adjustment [50.19438850112964]
  Regression tasks in computer vision are often formulated into classification by quantizing the target space into classes. The majority of training samples lie in a head range of target values, while a minority of samples span a usually larger tail range. We propose to construct hierarchical classifiers for solving imbalanced regression tasks. Our novel hierarchical classification adjustment (HCA) for imbalanced regression shows superior results on three diverse tasks.
  arXiv  Detail & Related papers  (2023-10-26T04:54:39Z)
• Calibrated and Sharp Uncertainties in Deep Learning via Density Estimation [10.209143402485406]
  This paper argues that calibration is important in practice and is easy to maintain.<n>We introduce a simple training procedure based on recalibration that yields calibrated models without sacrificing overall performance.
  arXiv  Detail & Related papers  (2021-12-14T06:19:05Z)
• Learning Prediction Intervals for Regression: Generalization and Calibration [12.576284277353606]
  We study the generation of prediction intervals in regression for uncertainty quantification. We use a general learning theory to characterize the optimality-feasibility tradeoff that encompasses Lipschitz continuity and VC-subgraph classes. We empirically demonstrate the strengths of our interval generation and calibration algorithms in terms of testing performances compared to existing benchmarks.
  arXiv  Detail & Related papers  (2021-02-26T17:55:30Z)
• Calibration of Neural Networks using Splines [51.42640515410253]
  Measuring calibration error amounts to comparing two empirical distributions. We introduce a binning-free calibration measure inspired by the classical Kolmogorov-Smirnov (KS) statistical test. Our method consistently outperforms existing methods on KS error as well as other commonly used calibration measures.
  arXiv  Detail & Related papers  (2020-06-23T07:18:05Z)

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.