AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Reliable Confidence Alignment for Generalized Category Discovery

Jiawei Yu, Zijian Gao, Tianjiao Wan, Xuan Liu, Cheng Yang, Kele Xu

ICML 2026 · Label-efficient and robust learning › Recognition and long-tail · approach: Learning from unlabeled and positive-unlabeled data

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add RAC (Dirichlet certainty anchoring on labeled data) (ablation)SimGCD · CUBOld accuracy: 67.38 -> 71.58Improved
add RAC (ablation)SimGCD · CIFAR100All accuracy: 79.43 -> 82.46Improved
add CCA on top of RAC (ablation)SimGCD · CUBOld accuracy: 71.58 -> 72.38Improved
add CCA on top of RAC (ablation)SimGCD · CIFAR100New accuracy: 82.65 -> 82.18Hurt
add CCA on top of RAC (ablation)SimGCD · CIFAR100All accuracy: 82.46 -> 82.85Improved
add RAC (ablation)AFGCD · CUBAll accuracy: 69.66 -> 70.73Improved
add CCA on top of RAC (ablation)AFGCD · CUBAll accuracy: 70.73 -> 71.20Improved
use Exp(Tanh) evidence instead of Softplus (ablation)RCA on SimGCD · FGVC-AircraftAll accuracy: 55.87 -> 56.79Improved
use Exp(Tanh) evidence instead of Exp (ablation)RCA on SimGCD · Stanford CarsAll accuracy: 55.26 -> 55.55Improved
increase RAC weight lambda1 from 0.1 to 0.7 (ablation)RCA · CUB and FGVC-AircraftNew accuracyHurt
vary CCA weight lambda2 from 0.1 to 0.7 (ablation)RCA · CUBAll accuracy: varies by only 0.1%No clear effect
add RCA to SimGCD on CIFAR10SimGCD · CIFAR10All accuracy: 97.0 -> 97.1No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy