NeurIPS 2026
Learning Multimodal One-step Flow Policy via Value-weighted Optimal Transport
OptiFlow uses value-weighted optimal transport to train efficient one-step flow policies that preserve high-value, multimodal behavior in offline reinforcement learning.