Icml_2026

Our works:

• Sharp description of local minima in the loss landscape of high-dimensional two-layer ReLU neural networks
• A Noise Sensitivity Exponent Controls Large Statistical-to-Computational Gaps in Single- and Multi-Index Models (Spotlight)
• A Random Matrix Theory of Masked Self-Supervised Regression
• On the existence of consistent adversarial attacks in high-dimensional linear classification (Spotlight)

were accepted for ICML 2026, two of them as spotlights! Come see the talks and posters in Seoul!