Socratic-MCTS: Test-Time Visual Reasoning by Asking the Right Questions
This addresses the challenge of enhancing existing deployed models for visual reasoning tasks, offering an incremental improvement over standard methods.
The paper tackled the problem of enabling non-reasoning vision-language models to perform long-form visual reasoning without additional training, by using a Monte Carlo Tree Search algorithm to inject subquestion-subanswer pairs, resulting in a 2% overall improvement on MMMU-PRO and a 9% gain in Liberal Arts.
Recent research in vision-language models (VLMs) has centered around the possibility of equipping them with implicit long-form chain-of-thought reasoning -- akin to the success observed in language models -- via distillation and reinforcement learning. But what about the non-reasoning models already trained and deployed across the internet? Should we simply abandon them, or is there hope for a search mechanism that can elicit hidden knowledge and induce long reasoning traces -- without any additional training or supervision? In this paper, we explore this possibility using a Monte Carlo Tree Search (MCTS)-inspired algorithm, which injects subquestion-subanswer pairs into the model's output stream. We show that framing reasoning as a search process -- where subquestions act as latent decisions within a broader inference trajectory -- helps the model "connect the dots" between fragmented knowledge and produce extended reasoning traces in non-reasoning models. We evaluate our method across three benchmarks and observe consistent improvements. Notably, our approach yields a 2% overall improvement on MMMU-PRO, including a significant 9% gain in Liberal Arts.