Marika NiihoriThe International Conference on Learning Representations (ICLR) 2026, held in Rio de Janeiro, Brazil from April 23 – 27, brought together the global machine learning community for one of the year’s leading events in the field.
It was a fantastic week of discussions and new ideas and one in which our group was recognised with two workshop awards: an Outstanding Paper Award and a Best Paper Award.
The Outstanding Paper Award (Workshop on LLM Reasoning) was awarded to Learning Reasoning Reward Models from Expert Demonstration via Inverse Reinforcement Learning. This work, led by Claudio Fanconi and Nicolás Astorga, introduces a framework for learning reward models that guide and evaluate reasoning processes in AI systems, enabling more robust and self-reflective decision-making.
The Best Paper Award (Agentic AI in the Wild: From Hallucinations to Reliable Autonomy) was awarded to GLEAN: Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification. Led by Yichi Zhang, a visiting PhD student in the group, with contributions from Dr Nabeel Seedat, an alumnus of the lab, this work proposes a method for verifying AI outputs by accumulating evidence grounded in domain-specific guidelines. The approach is particularly relevant in applications where reliability and accountability are critical.
Together, these works address two central challenges in modern AI: improving the reasoning capabilities of models and ensuring that their outputs can be trusted in real-world deployment. As AI systems are increasingly applied across domains such as bioscience discovery, healthcare, and education, such advances are essential for building robust and deployable systems.
The awarded papers can be accessed here:
- Outstanding Paper Award: Learning Reasoning Reward Models from Expert Demonstration via Inverse Reinforcement Learning: https://arxiv.org/abs/2510.01857
- Best Paper Award: GLEAN: Guideline-Grounded Evidence Accumulation for High-Stakes Agent Verification: https://arxiv.org/abs/2603.02798
