通过分析ViT各层输出概率变化，实现高效错误检测。
LogitDynamics: Reliable ViT Error Detection from Layerwise Logit Trajectories
- 在ViT中间层添加轻量头，追踪类别证据演化轨迹。
- 跨数据集测试中AUCPR优于或持平基线，且泛化能力强。
- 适合部署阶段的置信度评估，计算开销极低。
可靠的信心估计对视觉模型部署至关重要。本文研究误差预测：仅用单次前向传播信号判断图像分类器输出是否正确。受大语言模型内部信号幻觉检测启发，探究视觉变换器（ViTs）中是否存在类似深度层面信号。提出一种简单方法，建模类别证据在各层间的演化过程。通过在最后L层附加轻量级线性头，提取预测类别及其前K个竞争者的对数几率，以及顶层类别不稳定性统计特征。使用这些特征训练线性探测器预测错误指示符。在多个数据集上，该方法的AUCPR优于或持平基线，展现出更强的跨数据集泛化能力，且额外计算开销极小。
原文摘要 · Abstract (English)
Reliable confidence estimation is critical when deploying vision models. We study error prediction: determining whether an image classifier's output is correct using only signals from a single forward pass. Motivated by internal-signal hallucination detection in large language models, we investigate whether similar depth-wise signals exist in Vision Transformers (ViTs). We propose a simple method that models how class evidence evolves across layers. By attaching lightweight linear heads to intermediate layers, we extract features from the last L layers that capture both the logits of the predicted class and its top-K competitors, as well as statistics describing instability of top-ranked classes across depth. A linear probe trained on these features predicts the error indicator. Across datasets, our method improves or matches AUCPR over baselines and shows stronger cross-dataset generalization while requiring minimal additional computation.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。