08:15 - 08:20 Brief Introduction Irina Rish (CERC-AAI Lab/UdeM/Mila)
08:20 - 08:30 Introduction & Overview: KAIST Research (Slides), Jong Chul Ye (KAIST)
08:30 - 08:45 From Geometry to Worlds and Actions: Repurposing Geometric Foundation Models for 3D World Generation and Policy Learning, Seungryong Kim (KAIST)
08:45 - 09:00 Feed-forward 3D reconstruction and novel view synthesis, Eunbyung Park (Yonsei University)
09:00 - 09:15 One Step Video Editing with One Step Diffusion Models, Gyeong-Moon Park (Korea University)
09:15 - 09:30 coffee break
09:30 - 09:45 Advancing Data Attribution with Influence Functions, Dongwoo Kim (Postech)
10:00 - 10:15 Temporal Recurrence Favors Fewer Layers (Slides), Ivan Anokhin (Mila)
10:15 - 10:30 Learning Complex Behaviors from Offline Data, Jongmin Lee (Yonsei University)
10:30 - 10:45 On the Limits of Test-Time Compute for Self-Evaluation, Anirudh Buvanesh (Mila)
10:55- 11:10 Target-Aware Direct Preference Optimization for Longitudinal Post-Operative Tumor MRI Generation (Slides), Tae-Eui Kam (Korea University)
11:10 - 11:25 Towards Interactive and Reasoning-Capable Time-Series Foundation Models, Changhee Lee (Korea University)
11:25 - 11:40 Characterizing Latent Reasoning Dynamics in LLMs, Prateek Humane (Mila)
11:40 - 11:50 Concluding Remarks