--- title: "Fei Huang — papers, page 2 — Elamur Research" description: "Fei Huang, Alibaba Group: 45 papers on Elamur Research (2015–2026), h-index 43 — affiliations, topics, co-authors and who cites them in…" canonical: "https://research.elamur.ai/free/en/authors/fei-huang/page/2/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/authors/fei-huang/page/2/) [home](https://md.research.elamur.ai/free/en/) / [authors](https://md.research.elamur.ai/free/en/authors/) / [Fei Huang](https://md.research.elamur.ai/free/en/authors/fei-huang/) / page 2 # Fei Huang · page 2 45 papers in our library · 2015–2026 · page 2 of 2 [The profile, affiliations and co-authors →](https://md.research.elamur.ai/free/en/authors/fei-huang/) ## Papers The most cited first (OpenAlex and our library). [← prev](https://md.research.elamur.ai/free/en/authors/fei-huang/) [1](https://md.research.elamur.ai/free/en/authors/fei-huang/) **2** 26. **[AI Hospital: Benchmarking Large Language Models in a Multi-agent Medical Interaction Simulator](https://md.research.elamur.ai/free/en/papers/fan-2025-ai-hospital-benchmarking-large-language/)** *(2025 · 36 citations)* Medical QA benchmarks give a model the whole case at once. Real diagnosis is different: a clinician must gather facts step by step through conversation and tests. To measure this skill, the authors built AI Hospital, a multi-agent simulator. An LLM under test plays the Doctor, while GPT-3.5 drives two non-player… 27. **[Qwen3Guard Technical Report](https://md.research.elamur.ai/free/en/papers/zhao-2025-qwen3guard-technical-report/)** *(2025 · 33 citations)* Guardrail classifiers that screen LLM traffic usually produce only a binary safe/unsafe verdict, which different safety policies interpret differently. 28. **[ReSum: Unlocking Long-Horizon Search Intelligence via Context Summarization](https://md.research.elamur.ai/free/en/papers/wu-2025-resum-unlocking-long-horizon-search/)** *(2025 · 26 citations)* Web agents built on LLMs often fail on hard multi-hop questions. The ReAct loop keeps adding every search result to the prompt, so the context window fills up before the agent finds the answer. Earlier fixes such as MEM1 or MemAgent change the model's architecture and require full retraining. ReSum avoids this with a… 29. **[VRAG-RL: Empower Vision-Perception-Based RAG for Visually Rich Information Understanding via Iterative Reasoning with Reinforcement Learning](https://md.research.elamur.ai/free/en/papers/wang-2025-vrag-rl-empower-vision-perception/)** *(2025 · 23 citations)* Answering questions over visually rich documents such as slides, reports, charts and tables is difficult for standard RAG. 30. **[IOPO: Empowering LLMs with Complex Instruction Following via Input-Output Preference Optimization](https://md.research.elamur.ai/free/en/papers/zhang-2025-iopo-empowering-llms-complex-instruction/)** *(2025 · 14 citations)* Existing alignment methods such as DPO compare a preferred and a dispreferred response to the same prompt. When a prompt carries many constraints, this makes it hard for a model to notice which constraints matter. Training resources for this setting are also scarce. The authors address both gaps. 31. **[OmniThink: Expanding Knowledge Boundaries in Machine Writing through Thinking](https://md.research.elamur.ai/free/en/papers/xi-2025-omnithink-expanding-knowledge-boundaries-machine/)** *(2025 · 14 citations)* LLM-based long-form writing usually depends on retrieval-augmented generation, but standard retrieval tends to return shallow, repetitive, and unoriginal material. 32. **[QwenLong-L1: Towards Long-Context Large Reasoning Models with Reinforcement Learning](https://md.research.elamur.ai/free/en/papers/wan-2025-qwenlong-l1-long-context/)** *(2025 · 13 citations)* Reinforcement learning has made large reasoning models (LRMs) much better at problems with short inputs. It is still unclear how to train them to reason over very long documents, potentially around 120K tokens. The authors frame this as "long-context reasoning RL," where the model must first find and ground the… 33. **[DeepSolution: Boosting Complex Engineering Solution Design via Tree-based Exploration and Bi-point Thinking](https://md.research.elamur.ai/free/en/papers/li-2025-deepsolution-boosting-complex-engineering-solution/)** *(2025 · 10 citations)* Research on retrieval-augmented generation (RAG) has mostly targeted multi-hop and long-form question answering. 34. **[QwenLong-L1.5: Post-Training Recipe for Long-Context Reasoning and Memory Management](https://md.research.elamur.ai/free/en/papers/shen-2025-qwenlong-l1-5-post/)** *(2025 · 10 citations)* Research on long contexts has mostly targeted pre-training and architecture, so post-training lags behind. 35. **[RL-PLUS: Countering Capability Boundary Collapse of LLMs in Reinforcement Learning with Hybrid-policy Optimization](https://md.research.elamur.ai/free/en/papers/dong-2026-rl-plus-countering-capability-boundary/)** *(2026 · 6 citations)* RL-PLUS targets a known weakness of Reinforcement Learning with Verifiable Reward (RLVR). Standard on-policy training such as GRPO raises pass@1 but often lowers pass@k at large k. In other words, the tuned model solves fewer distinct problems than its base model, which the authors call capability boundary collapse.… 36. **[EIFBENCH: Extremely Complex Instruction Following Benchmark for Large Language Models](https://md.research.elamur.ai/free/en/papers/zou-2025-eifbench-extremely-complex-instruction-following/)** *(2025 · 5 citations)* Most instruction-following benchmarks test either one task with several constraints or several tasks with one constraint each. 37. **[Writing-RL: Advancing Long-form Writing via Adaptive Curriculum Reinforcement Learning](https://md.research.elamur.ai/free/en/papers/lei-2026-writing-rl-advancing-long-form/)** *(2026 · 4 citations)* Large language models can now produce long articles, but further gains are hard to get. Supervised fine-tuning stalls because it can only imitate teacher or human outputs, and the RL recipes that work for math and code need verifiable answers that open-ended writing lacks. The authors propose Writing-RL, a… 38. **[EvoRoute: Experience-Driven Self-Routing LLM Agent Systems](https://md.research.elamur.ai/free/en/papers/zhang-2026-evoroute-experience-driven-self-routing/)** *(2026 · 2 citations)* Multi-model agentic systems for deep research perform well but often run slowly and cost a lot. The authors frame this as the Agent System Trilemma : it is hard to get high accuracy, low monetary cost and fast completion at the same time. Their solution, EvoRoute , chooses an LLM backbone for every individual step of… 39. **[MUSEG: Reinforcing Video Temporal Understanding via Timestamp-Aware Multi-Segment Grounding](https://md.research.elamur.ai/free/en/papers/luo-2026-museg-reinforcing-video-temporal-understanding/)** *(2026 · 1 citation)* Multimodal LLMs remain weak at fine-grained temporal reasoning in video, such as locating when events happen. 40. **[Scaling External Knowledge Input Beyond Context Windows of LLMs via Multi-Agent Collaboration](https://md.research.elamur.ai/free/en/papers/liu-2026-scaling-external-knowledge-input-beyond/)** *(2026 · 1 citation)* Large language models can use more retrieved evidence to solve harder tasks, but their context window caps how much they can take in. 41. **[A Novel Matching Paradigm: Unified Generative and Discriminative LLM with Prompt Compression for Relevance Learning](https://md.research.elamur.ai/free/en/papers/zhao-2026-novel-matching-paradigm-unified-generative/)** *(2026)* The paper addresses query–landing-page relevance in industrial search advertising. In this setting, LLMs usually act only as feature extractors, and two-tower or single-tower matchers return bare scores with no explanation. The authors, from Baidu Search Ads and Xi'an Jiaotong University, propose UGD (Unified… 42. **[BrowseConf: Confidence-Guided Test-Time Scaling for Web Agents](https://md.research.elamur.ai/free/en/papers/ou-2026-browseconf-confidence-guided-test-time/)** *(2026)* The paper asks whether LLM-based web search agents can meaningfully report how sure they are after long chains of browsing and reasoning. 43. **[Experience-driven Multi-turn Reinforcement Learning for GUI Agents](https://md.research.elamur.ai/free/en/papers/lu-2026-experience-driven-multi-turn-reinforcement/)** *(2026)* Training vision-based GUI agents for long, multi-step tasks involves a trade-off. Single-turn RL methods such as GRPO learn from expert histories, but at deployment the agent must rely on its own past actions, so training and real use differ. Online multi-turn RL closes this gap, but it brings sparse rewards, high… 44. **[Format-Adapter: Improving Reasoning Capability of LLMs by Adapting Suitable Format](https://md.research.elamur.ai/free/en/papers/wang-2026-format-adapter-improving-reasoning-capability/)** *(2026)* Parallel test-time scaling (TTS) usually samples many answers using a single reasoning format chosen by a human. 45. **[Qwen-AgentWorld: Language World Models for General Agents](https://md.research.elamur.ai/free/en/papers/zuo-2026-qwen-agentworld-language-world/)** *(2026)* LLM agent research has mostly focused on the policy, which picks actions from states. The complementary piece is a world model that predicts the next state from a state and an action, and agents working in text-based environments still lack a general one. The Qwen Team builds such a model, Qwen-AgentWorld. It… [← prev](https://md.research.elamur.ai/free/en/authors/fei-huang/) [1](https://md.research.elamur.ai/free/en/authors/fei-huang/) **2**