Dong Yu · page 2
Papers
- WebAggregator: Enhancing Compositional Reasoning Capabilities of Deep Research Agent Foundation Models (2026 · 1 citation)
Deep Research agents need more than retrieval. They must combine scattered, heterogeneous evidence through multi-step reasoning. Existing training and evaluation resources mostly reward locating a single entity, so agents learn to call tools often instead of reasoning carefully. - WebRollback: Enhancing Web Agents with Explicit Rollback Mechanisms (2026 · 1 citation)
Most LLM-based web agents navigate greedily in a single direction. On live, changing websites they can enter a wrong state and waste many steps before recovering, if they recover at all. Parallel search methods are costly here because each branch needs its own browser environment. Best-first search, as in Koh et al.… - Beyond Detection: Evaluating Fallacy Awareness of LLMs in Interactive Scenarios (2026)
LLMs can usually name and classify a fallacy when explicitly asked to, but they often fail to notice flawed reasoning during conversation. - Crossing the Reward Bridge: Expanding Reinforcement Learning with Verifiable Rewards Across Diverse Domains (2026)
RLVR works well for math and code, where a simple rule can check whether an answer is right. Such checkers fail on free-form answers, and the authors show that rule-based rewards can even get worse as more mixed-domain training data is added. The paper replaces the rule with a generative verifier. An LLM compares the… - Measure Twice, Click Once: Co-evolving Proposer and Visual Critic via Reinforcement Learning for GUI Grounding (2026)
GUI grounding means turning a natural-language instruction into exact click coordinates on a screenshot. MLLMs find it hard on high-resolution, crowded interfaces: they often understand the request but miss the precise pixel location. Drawing several samples raises Pass@k. However, clustering-style self-consistency… - SPAGBias: Uncovering and Tracing Structured Spatial Gender Bias in Large Language Models (2026)
LLMs are increasingly used in urban planning and other spatial tasks. Feminist geography holds that everyday places carry gendered meanings, so these models may reproduce or amplify that hierarchy. The authors present SPAGBias, described as the first systematic approach to quantifying how LLMs associate gender with…
This page in Markdown → https://md.research.elamur.ai/free/en/authors/dong-yu/page/2/