2026 01 19__iclr

Three new papers were accepted to ICLR2026!

• Discrete Adjoint Matching: Extending adjoint matching to discrete state spaces.
• ReFORM: Reflected Flows for On-support Offline RL via Noise Manipulation: Improving offline RL using support constraints instead of regularizing statistical divergence.
• Solving Parameter-Robust Avoid Problems with Unknown Feasibility using Reinforcement Learning: A step towards solving for safe policies that are robust to parameter sets that may include infeasible parameters.

I am unable to travel to Rio de Janeiro 🇧🇷 to present these works in person, but my co-authors will be there to present these works.