--- title: "Pengfei Liu — papers, page 2 · Shenyang Institute of Automation" description: "Pengfei Liu, Shenyang Institute of Automation: 36 papers on Elamur Research (2015–2026), h-index 31 — affiliations, topics, co-authors and…" canonical: "https://research.elamur.ai/free/en/authors/pengfei-liu/page/2/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/authors/pengfei-liu/page/2/) [home](https://md.research.elamur.ai/free/en/) / [authors](https://md.research.elamur.ai/free/en/authors/) / [Pengfei Liu](https://md.research.elamur.ai/free/en/authors/pengfei-liu/) / page 2 # Pengfei Liu · page 2 36 papers in our library · 2015–2026 · page 2 of 2 [The profile, affiliations and co-authors →](https://md.research.elamur.ai/free/en/authors/pengfei-liu/) ## Papers The most cited first (OpenAlex and our library). [← prev](https://md.research.elamur.ai/free/en/authors/pengfei-liu/) [1](https://md.research.elamur.ai/free/en/authors/pengfei-liu/) **2** 26. **[O1 Replication Journey: A Strategic Progress Report — Part 1](https://md.research.elamur.ai/free/en/papers/qin-2024-o1-replication-journey-strategic-progress/)** *(2024 · 86 citations)* OpenAI's O1 model showed strong multi-step reasoning, but its training recipe, data and full outputs were never disclosed. 27. **[Evaluating Mathematical Reasoning Beyond Accuracy](https://md.research.elamur.ai/free/en/papers/xia-2024-evaluating-mathematical-reasoning-accuracy/)** *(2024 · 85 citations)* Most benchmarks for LLM math reasoning check only whether the final answer is right. This can hide flawed logic or wasted steps inside a solution. The authors propose REASONEVAL, which scores every step of a solution on two dimensions: 28. **[Benchmarking Benchmark Leakage in Large Language Models](https://md.research.elamur.ai/free/en/papers/xu-2024-benchmarking-benchmark-leakage-large-language/)** *(2024 · 73 citations)* Large language models are often trained on undisclosed corpora that may include public benchmark data, sometimes on purpose. 29. **[ToRL: Scaling Tool-Integrated RL](https://md.research.elamur.ai/free/en/papers/li-2025-torl-scaling-tool-integrated-rl/)** *(2025 · 61 citations)* Language models that reason only in natural language often fail at exact arithmetic, equation solving, and exhaustive enumeration. 30. **[Progress or Regress? Self-Improvement Reversal in Post-training](https://md.research.elamur.ai/free/en/papers/wu-2024-progress-regress-self-improvement-reversal/)** *(2024 · 10 citations)* LLMs are often made to "self-improve" by repeatedly fine-tuning them on their own generated solutions, with success usually judged only by pass@1 on reasoning benchmarks. 31. **[Towards Dynamic Theory of Mind: Evaluating LLM Adaptation to Temporal Evolution of Human States](https://md.research.elamur.ai/free/en/papers/xiao-2025-dynamic-theory-mind-evaluating-llm/)** *(2025 · 8 citations)* Most Theory of Mind (ToM) benchmarks for LLMs test mental states at a single moment. In real social interaction, however, people's beliefs, feelings and goals change over time. The authors address this gap with DynToM, a benchmark that checks whether models can follow how a character's mental state develops across a… 32. **[Halu-J: Critique-Based Hallucination Judge](https://md.research.elamur.ai/free/en/papers/wang-2024-halu-j-critique-based-hallucination/)** *(2024 · 8 citations)* LLMs often produce factually wrong statements. Retrieval-based detectors usually reduce the check to a bare label. They give little justification, can be thrown off by off-topic retrieved passages, and weigh all evidence equally. Halu-J addresses this with an open judge model of 7 billion parameters. Given a claim… 33. **[DiagnosisArena: Benchmarking Diagnostic Reasoning for Large Language Models](https://md.research.elamur.ai/free/en/papers/zhu-2026-diagnosisarena-benchmarking-diagnostic-reasoning-large/)** *(2026 · 7 citations)* Popular medical LLM benchmarks such as MedQA and MMLU are nearly saturated, with leading models scoring above 90%. 34. **[SepSeq: A Training-Free Framework for Long Numerical Sequence Processing in LLMs](https://md.research.elamur.ai/free/en/papers/sun-2026-sepseq-training-free-framework-long/)** *(2026 · 1 citation)* Modern LLMs advertise very long context windows, yet they handle long sequences of numbers poorly. Tasks such as counting, filtering, or spotting trends need exact access to every value. The authors trace this failure to attention dispersion: as inputs grow, Softmax spreads attention over many similar numeric tokens… 35. **[Test-Time Adaptation of an Offline Multimodal Foundation Model for Simultaneous Speech Translation](https://md.research.elamur.ai/free/en/papers/xing-2026-test-time-adaptation-offline-multimodal/)** *(2026 · 1 citation)* The paper addresses end-to-end simultaneous speech-to-text translation (SimulST), where systems usually need complex architectures and dedicated training. 36. **[SciPedia: Unlocking the Value of Scientific Data for Pre-training](https://md.research.elamur.ai/free/en/papers/qin-2026-scipedia-unlocking-value-scientific-data/)** *(2026)* Academic books and research papers are among the richest sources of high-quality text. Open LLM pre-training seldom uses them because public scientific corpora are mostly web-filtered or narrow, and much of the literature is locked inside PDFs. The authors collect books and papers from original sources and process… [← prev](https://md.research.elamur.ai/free/en/authors/pengfei-liu/) [1](https://md.research.elamur.ai/free/en/authors/pengfei-liu/) **2**