FedCare delivers the first visual pipeline that pinpoints, classifies and mitigates FL failures in real time, cutting convergence rounds and boosting accuracy across tested attacks without sacrificing privacy.

DVRE cuts MaxSAT encoding size offline and clauses online, delivering the fastest complete MBD on ISCAS-85 for both single- and multi-observation cases without extra data.

Research team debuts the first visual pre-training paradigm tailored for CTR prediction, lifting Taobao GMV by 0.88% (p < 0.01) via user-intent contrastive learning that mines interest-centric image signals.

LAMDA-SSL toolkit delivers the first unified benchmarks and robust algorithms that safely exploit unlabeled data despite label, feature and distribution mismatches, beating vanilla supervised baselines in open-world SSL.

RetinaDA unites six public fundus sets into a 512 × 512 macula-centered benchmark with built-in domain gaps, enabling vessel-segmentation models to generalize across devices without extra annotation cost.

FPMCO decomposes multi-constraint RL into KL-projection sub-problems, achieving higher reward with lower computing than second-order rivals on the new SCIG robotics benchmark.