When Probing Accuracy Saturates, Fragility Resolves: A Complementary Metric for LLM Pre-Training Analysis

• URL: http://arxiv.org/abs/2606.11375v1
• Date: Tue, 09 Jun 2026 19:00:23 GMT
• Title: When Probing Accuracy Saturates, Fragility Resolves: A Complementary Metric for LLM Pre-Training Analysis
• Abstract summary: Linear probing declares a property "encoded" when a classifier on hidden states achieves high accuracy.<n>Standard linear probing saturates within the first few thousand steps, leaving most of training invisible to the instrument.<n>We introduce fragility, a complementary per-layer metric defined as the activation-noise level at which probe accuracy collapses.
• Score: 1.0829694003408499
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Standard linear probing declares a property "encoded" when a classifier on hidden states achieves high accuracy. The protocol works well on a snapshot but breaks across pre-training: probe accuracy saturates within the first few thousand steps, leaving most of training invisible to the instrument. We introduce fragility, a complementary per-layer metric defined as the activation-noise level at which probe accuracy collapses. Fragility is sensitive to both the margin of separability and the redundancy of representation, both of which keep evolving long after accuracy plateaus. Applied to open-checkpoint language models, fragility recovers structure that accuracy alone cannot see. Moralized representations emerge along a lexical $\to$ compositional gradient: lexical moral detection first, compositional moral encoding later. Because probe accuracy on its own tracks how lexically separable a dataset is, we establish the compositional encoding directly, by showing it transfers across construction types that share no contrast tokens. A layer-depth robustness gradient develops monotonically across training while accuracy stays flat. And matched fine-tuning corpora that produce identical probing accuracy leave distinct fragility fingerprints, showing that data curation reshapes probe robustness without changing probe accuracy. In every comparison we test, where probing accuracy returns a flat answer, fragility returns a structured one.

Related papers

• Subtype Robustness Is Not Just Accuracy: Calibration Under Unseen Subtype Shift [1.2085879670633297]
  Subtype robustness asks whether a model keeps the correct coarse prediction when test examples come from fine-grained subtypes absent from training but still inside a known coarse category.<n>We present the first systematic study of the question across ImageNet, BREEDS, iNaturalist and CIFAR-100 with five architectures.
  arXiv  Detail & Related papers  (2026-08-02T01:49:59Z)
• Recovering Input Text from Hidden States: Study of Gradient-Based Inversion of Decoder-Only Language Models [0.0]
  We study the hidden-state inversion problem as a continuous embedding-space optimisation.<n>We analyse which tokens break the reconstructions and find a sharp categorical asymmetry.<n>Last-layer hidden states of GPT-2 are as sensitive as original text.
  arXiv  Detail & Related papers  (2026-07-01T12:18:12Z)
• Measuring Temporal Linguistic Emergence in Diffusion Language Models [0.0]
  Diffusion language models expose an explicit denoising trajectory.<n>We study three independent 32-step runs of LLaDA-8B-Base on WikiText-103 text.
  arXiv  Detail & Related papers  (2026-04-25T10:17:47Z)
• LRANet++: Low-Rank Approximation Network for Accurate and Efficient Text Spotting [118.93173826110815]
  We propose a novel parameterized text shape method based on low-rank approximation for precise detection.<n>By exploiting the inherent shape correlation among different text contours, our method achieves consistency and compactness in shape representation.<n>We integrate the enhanced detection module with a lightweight recognition branch to form an end-to-end text spotting framework, termed LRANet++.
  arXiv  Detail & Related papers  (2025-11-08T03:08:03Z)
• Ambiguity-aware Point Cloud Segmentation by Adaptive Margin Contrastive Learning [65.94127546086156]
  We propose an adaptive margin contrastive learning method for semantic segmentation on point clouds.<n>We first design AMContrast3D, a method comprising contrastive learning into an ambiguity estimation framework.<n>Inspired by the insight of joint training, we propose AMContrast3D++ integrating with two branches trained in parallel.
  arXiv  Detail & Related papers  (2025-07-09T07:00:32Z)
• Ethicist: Targeted Training Data Extraction Through Loss Smoothed Soft Prompting and Calibrated Confidence Estimation [56.57532238195446]
  We propose a method named Ethicist for targeted training data extraction. To elicit memorization, we tune soft prompt embeddings while keeping the model fixed. We show that Ethicist significantly improves the extraction performance on a recently proposed public benchmark.
  arXiv  Detail & Related papers  (2023-07-10T08:03:41Z)
• Improving the Accuracy-Robustness Trade-Off of Classifiers via Adaptive Smoothing [9.637143119088426]
  We show that a robust base classifier's confidence difference for correct and incorrect examples is the key to this improvement. We adapt an adversarial input detector into a mixing network that adaptively adjusts the mixture of the two base models. The proposed flexible method, termed "adaptive smoothing", can work in conjunction with existing or even future methods that improve clean accuracy, robustness, or adversary detection.
  arXiv  Detail & Related papers  (2023-01-29T22:05:28Z)
• Confidence Calibration for Intent Detection via Hyperspherical Space and Rebalanced Accuracy-Uncertainty Loss [17.26964140836123]
  In some scenarios, users do not only care about the accuracy but also the confidence of model. We propose a model using the hyperspherical space and rebalanced accuracy-uncertainty loss. Our model outperforms the existing calibration methods and achieves a significant improvement on the calibration metric.
  arXiv  Detail & Related papers  (2022-03-17T12:01:33Z)
• On the (Un-)Avoidability of Adversarial Examples [4.822598110892847]
  adversarial examples in deep learning models have caused substantial concern over their reliability. We provide a framework for determining whether a model's label change under small perturbation is justified. We prove that our adaptive data-augmentation maintains consistency of 1-nearest neighbor classification under deterministic labels.
  arXiv  Detail & Related papers  (2021-06-24T21:35:25Z)
• Information-Theoretic Probing with Minimum Description Length [74.29846942213445]
  We propose an alternative to the standard probes, information-theoretic probing with minimum description length (MDL) With MDL probing, training a probe to predict labels is recast as teaching it to effectively transmit the data. We show that these methods agree in results and are more informative and stable than the standard probes.
  arXiv  Detail & Related papers  (2020-03-27T09:35:38Z)
• Post-Estimation Smoothing: A Simple Baseline for Learning with Side Information [102.18616819054368]
  We propose a post-estimation smoothing operator as a fast and effective method for incorporating structural index data into prediction. Because the smoothing step is separate from the original predictor, it applies to a broad class of machine learning tasks. Our experiments on large scale spatial and temporal datasets highlight the speed and accuracy of post-estimation smoothing in practice.
  arXiv  Detail & Related papers  (2020-03-12T18:04:20Z)
• A Closer Look at Accuracy vs. Robustness [94.2226357646813]
  Current methods for training robust networks lead to a drop in test accuracy. We show that real image datasets are actually separated. We conclude that achieving robustness and accuracy in practice may require using methods that impose local Lipschitzness.
  arXiv  Detail & Related papers  (2020-03-05T07:09:32Z)

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.