SCICON: Contrastive Decoding for Scientific Figure QA
Discover SCICON, a novel contrastive decoding method that reduces bias in scientific figure multiple-choice QA by prioritizing visual evidence over text.
Autonomous Agent Reveals Transferable Molecular Transformer Designs
Discover how autonomous agents optimize molecular transformer designs and transfer innovations across SMILES, proteins, and natural language models.
HeteroHub: Data Framework for Multi-Agent AI Systems
Discover HeteroHub, a data management framework enabling efficient coordination and real-time control of heterogeneous multi-embodied AI agents.
SARL: Label-Free Reinforcement Learning via Reasoning Topology
Discover SARL, a label-free reinforcement learning method that rewards reasoning structure for better generalization in complex tasks.
CARV Benchmark: Testing Analogical Reasoning in Multimodal LLMs
Explore CARV, a new benchmark assessing compositional analogical reasoning in multimodal large language models, revealing key AI challenges and insights.