MATA: A Trainable Hierarchical Automaton System for Multi-Agent Visual Reasoning
Preprint: MATA introduces a hierarchical automaton system for visual reasoning, using…
arxiv.org/abs/2601.19204 #AI #VisualReasoning #MultiAgent #MachineLearning #Preprint #Arxiv #ScienceNews
MATA: A Trainable Hierarchical Automaton System for Multi-Agent...
Recent vision-language models have strong perceptual ability but their implicit reasoning is hard to explain and easily generates hallucinations on complex queries. Compositional methods improve...
From arxiv.org 10
