arXiv:2605. 30188v2 Announce Type: replace-cross Abstract: Reliable probability estimates are critical in many machine learning applications, yet modern classifiers are often poorly calibrated.
By Eug\`ene Berta, David Holzm\"uller, Francis Bach, Michael I. Jordan
arXiv:2606. 28654v1 Announce Type: cross Abstract: Deep Neural Network (DNN) classifiers suffer from poor calibration when their softmax outputs (predictive confidence) deviate from the empirical likelihoods.
By Thiru Thillai Nadarasar Bahavan, Sachith Seneviratne, Saman Halgamuge
The paper introduces Learning Neighbor Regions (LNR), a coarse‑to‑fine framework for robust multi‑model fitting in noisy environments. LNR’s coarse module uses a neural network to analyze geometric features of minimum sets, selecting a small number of promising hypotheses to improve efficiency. The fine module encodes neighbor region features for each hypothesis, allowing simultaneous refinement and scoring while avoiding differentiation of the sampling process and solvers. Experiments on four classic tasks show state‑of‑the‑art performance and suggest easy adaptation to other robust fitting problems.
By Chang Nie, Guangming Wang, Zhe Liu, Hesheng Wang
The paper introduces SMART, a lightweight recalibration technique that adjusts logits based on the margin between the top two logits, called the logit gap. It uses a soft-binned Expected Calibration Error objective to balance bias and variance, enabling stable updates even with limited calibration data. Experiments across various datasets and architectures show SMART achieves state‑of‑the‑art calibration with fewer parameters than existing methods.
By Haolan Guo, Linwei Tao, Haoyang Luo, Minjing Dong, Chang Xu
The paper introduces EUA‑Cal, a calibration method that preserves the early‑training uncertainty of deep neural networks. By regularizing predictions and feature prototypes based on the early model, EUA‑Cal mitigates overconfidence while maintaining accuracy. Experiments on image classification and multiple‑choice question answering across eight models show EUA‑Cal outperforms existing calibration techniques.
By Yutong Xie, Jiawei Tang, Zhenglin Hua, Yuxiang Ma, Si Qin, Yaxin Hou, Hui Liu, Junhui Hou, Yuheng Jia
arXiv:2608.30230v1 Announce Type: new
Abstract: Test-time prompt tuning (TPT) has emerged as a powerful paradigm, refining prompts for each test sample via entropy minimization (EM) over multiple aug...
By Jungwon Choi, Hyeonseo Jang, Kibok Lee, Eunwoo Kim