AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Distributionally Robust Set Representation Learning Under Inference-Time Element Corruption

Yankai Chen, Hanrong Zhang, Bowei He, Philip S. Yu, Xue Liu

ICML 2026 · Learning theory and algorithms › Prediction and decision losses · approach: Distributionally robust optimization · 0 citations (OpenAlex)

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 SW-DRSO robust training instead of ERM (ablation)SW-DRSO set encoder · LiveJRecall@10 (overall): 72.58 -> 74.86Improved
add SW-DRSO robust training instead of ERM (ablation)SW-DRSO set encoder · LiveJ severeRecall@10: 70.96 -> 73.25Improved
add sliced-Wasserstein encoder f_sw instead of mean pooling (ablation)SW-DRSO · LiveJRecall@10 (overall): 69.55 -> 74.86Improved
add Wasserstein barycenter synthesis instead of random in-batch embeddings (ablation)SW-DRSO · LiveJRecall@10 (overall): 72.92 -> 74.86Improved
use sliced Wasserstein instead of Sinkhorn Wasserstein (ablation)SW-DRSO · LiveJtraining time per epoch (min): 21.5 -> 15.6Improved
use sliced Wasserstein instead of Sinkhorn Wasserstein (ablation)SW-DRSO · LiveJRecall@10 (overall): 74.81 -> 74.86No clear effect
use barycentric adversary instead of discrete adversary (ablation)SW-DRSO · LiveJRecall@10 (overall): 74.45 -> 74.86Improved
use barycentric adversary instead of discrete adversary (ablation)SW-DRSO · LiveJtraining time per epoch (min): 18.4 -> 15.6Improved
increase random combinatorial search rounds m from 2 to 10 (ablation)RCS · LiveJRecall@10: 69.45 -> 71.13; time 21.7 -> 27.4Mixed
use barycentric adversary instead of random combinatorial search (ablation)SW-DRSO · LiveJRecall@10: 71.13 -> 74.86Improved
use SW-DRSO instead of WDRODRO variant · LiveJRecall@10 (overall): 74.72 -> 74.86Improved
use SW-DRSO instead of KL-DRODRO variant · LiveJtraining time per epoch (min): 14.2 -> 15.6Hurt

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