RLER: A New Paradigm for Video Reasoning Combining Reinforcement Learning and Evidence Election

Introduction / Main Floor: RLER: A New Paradigm for Video Reasoning Combining Reinforcement Learning and Evidence Election

This paper proposes the RLER dual-paradigm framework, which trains models to generate structured evidence via reinforcement learning and then selects reliable answers through a training-free evidence weighted election mechanism. It achieves SOTA on 8 video reasoning benchmarks, with an average improvement of 6.3% and only requiring 3.1 candidates.