…We propose ROMA, an RL fine-tuning framework that modifies the optimization dynamics to reinforce reasoning against visual degradation while preserving clean-input performance. A dual-forward-pass strategy uses teacher forcing…
…Bringing Agentic Search into Visual Perception Published on May 12 Submitted by taesiri on May 13 Authors: , , , , , Abstract Researchers introduce WebEye, a benchmark for object localization requiring external knowledge resolution, and Pixel…
…To support this setting, we construct CLOC, a Cross-domain Large-scale Object Counting dataset that reorganizes diverse public data sources into a unified benchmark. CLOC covers six visual domains: General Scene…
…Bohan Lyu , , , Jiaru Zhang , Qixin Xu , , , , , , , , , Junlin Yang , , , , , , , , Abstract Current AI agents struggle to invent generalizable and scalable machine learning methods, relying more on engineering tuning than true method discovery, with performance…
…During inference, a fast capability loop updates agent expertise using trajectory-level feedback , while a slow meta-LLM-driven topology loop performs agents' birth-death operations on MAS, including edge edit , agent…
…Yao Du , , Abstract A distribution-aware reinforcement learning framework improves multimodal large language models' numerical regression performance on long-tailed distributions through batch-level comparison-based supervision. AI-generated summary Multimodal large…
…Hyeon Hwang , , Abstract ThinC framework enables mathematical problem solving where code serves as the primary reasoning mechanism instead of a verification tool, demonstrating superior performance on math benchmarks. AI-generated summary Tool…
…Shijue Huang , , , , , , , , , Abstract A visual-native agent harness with image bank reference protocol enables reusable intermediate visual evidence and closed-loop data generation that improves multimodal deep search performance across multiple benchmarks…
…Xingyuan Hua , , Abstract Agents use variational inference to evaluate exploratory actions and selectively explore only when uncertainty is high, improving performance on text-based and GUI-based benchmarks. AI-generated summary Recent…
…Ziyu Guo , , , Abstract ATLAS presents a visual reasoning framework that combines agentic operations and latent representations using functional tokens, enabling efficient training and improved performance on complex benchmarks. AI-generated summary Visual…