--- title: "Yang Liu — papers, page 2 — Elamur Research" description: "Yang Liu, Beijing University of Posts and Telecommunications: 76 papers on Elamur Research (2006–2026), h-index 1 — affiliations, topics…" canonical: "https://research.elamur.ai/free/en/authors/yang-liu/page/2/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/authors/yang-liu/page/2/) [home](https://md.research.elamur.ai/free/en/) / [authors](https://md.research.elamur.ai/free/en/authors/) / [Yang Liu](https://md.research.elamur.ai/free/en/authors/yang-liu/) / page 2 # Yang Liu · page 2 76 papers in our library · 2006–2026 · page 2 of 4 [The profile, affiliations and co-authors →](https://md.research.elamur.ai/free/en/authors/yang-liu/) ## Papers The most cited first (OpenAlex and our library). [← prev](https://md.research.elamur.ai/free/en/authors/yang-liu/) [1](https://md.research.elamur.ai/free/en/authors/yang-liu/) **2** [3](https://md.research.elamur.ai/free/en/authors/yang-liu/page/3/) [4](https://md.research.elamur.ai/free/en/authors/yang-liu/page/4/) [next →](https://md.research.elamur.ai/free/en/authors/yang-liu/page/3/) 26. **[Self-Knowledge Guided Retrieval Augmentation for Large Language Models](https://md.research.elamur.ai/free/en/papers/wang-2023-self-knowledge-guided-retrieval-augmentation/)** *(2023 · 135 citations)* Retrieval augmentation is often treated as universally helpful for large language models, but the authors show that retrieved passages can sometimes distract a model from answers it would otherwise get right. 27. **[Recognizing Implicit Discourse Relations via Repeated Reading: Neural Networks with Multi-Level Attention](https://md.research.elamur.ai/free/en/papers/liu-2016-recognizing-implicit-discourse-relations-repeated/)** *(2016 · 128 citations)* Implicit discourse relations, such as contrast or causality between two text spans (arguments) that lack an overt connective like "because", are hard to classify automatically. 28. **[Skywork-Reward: Bag of Tricks for Reward Modeling in LLMs](https://md.research.elamur.ai/free/en/papers/liu-2024-skywork-reward-bag-tricks/)** *(2024 · 112 citations)* Reward models turn human preferences into a training signal for LLM alignment. Their quality depends heavily on the preference data, and open-source preference datasets tend to be noisy or inconsistently labelled. The authors take a data-first approach. Using only public sources (HelpSteer2, OffsetBias, WildGuardMix… 29. **[Exploring Large Language Models for Communication Games: An Empirical Study on Werewolf](https://md.research.elamur.ai/free/en/papers/xu-2023-exploring-large-language-models-communication/)** *(2023 · 97 citations)* Communication games are incomplete-information games in which players must reason through natural language. 30. **[PLACES: Prompting Language Models for Social Conversation Synthesis](https://md.research.elamur.ai/free/en/papers/chen-2023-places-prompting-language-models-social/)** *(2023 · 90 citations)* Gathering good dialogue data through crowdsourcing is costly, and for some applications privacy or ethical constraints make it impossible. 31. **[Inference-Time Scaling for Generalist Reward Modeling](https://md.research.elamur.ai/free/en/papers/liu-2025-inference-time-scaling-generalist/)** *(2025 · 65 citations)* Reinforcement learning for LLMs depends on reliable reward signals. Outside verifiable areas like math or code, these are hard to obtain because the criteria vary and no ground truth exists. The authors ask whether a reward model (RM) can be trained so that giving it more inference compute yields better rewards on… 32. **[Samba: Simple Hybrid State Space Models for Efficient Unlimited Context Language Modeling](https://md.research.elamur.ai/free/en/papers/ren-2024-samba-simple-hybrid-state-space/)** *(2024 · 59 citations)* Language models struggle with very long inputs. Full attention scales quadratically with sequence length, while state space models (SSMs) are linear but recall specific details poorly. The authors propose Samba, a hybrid that stacks Mamba layers, SwiGLU MLPs and Sliding Window Attention (SWA, window 2048) in… 33. **[Skywork-Reward-V2: Scaling Preference Data Curation via Human-AI Synergy](https://md.research.elamur.ai/free/en/papers/liu-2025-skywork-reward-v2-scaling/)** *(2025 · 48 citations)* Open reward models used in RLHF are fragile. High RewardBench scores often fail to carry over to other evaluations, and tweaks to losses or architectures have brought little real progress. The authors argue that the bottleneck is the preference data itself, which is often narrow, synthetically labeled or poorly… 34. **[Learning Contextually Informed Representations for Linear-Time Discourse Parsing](https://md.research.elamur.ai/free/en/papers/liu-2017-learning-contextually-informed-representations-linear/)** *(2017 · 44 citations)* The paper addresses a trade-off in RST discourse parsing. Cubic-time CKY-style parsers reach top accuracy but are slow on long documents. Linear-time parsers are fast but usually depend on local, hand-engineered features. The authors present CIDER, a greedy, bottom-up, two-stage parser. It first builds a discourse… 35. **[MACSum: Controllable Summarization with Mixed Attributes](https://md.research.elamur.ai/free/en/papers/zhang-2023-macsum-controllable-summarization-mixed-attributes/)** *(2023 · 37 citations)* Readers often want summaries tailored to them, for example short ones, ones about a specific topic, or ones that stay close to the source wording. 36. **[Vision-Language Models Can Self-Improve Reasoning via Reflection](https://md.research.elamur.ai/free/en/papers/cheng-2025-vision-language-models-can-self/)** *(2025 · 35 citations)* Open multimodal LLMs reason poorly with chain-of-thought (CoT), and annotated multimodal rationales are expensive to collect at scale. 37. **[Enhancing Multilingual Capabilities of Large Language Models through Self-Distillation from Resource-Rich Languages](https://md.research.elamur.ai/free/en/papers/zhang-2024-enhancing-multilingual-capabilities-large-language/)** *(2024 · 30 citations)* LLMs trained mostly on English handle most other languages much worse. The usual remedy is to machine-translate instruction data into the target language and fine-tune on it. The authors argue this approach has three weaknesses: it ignores what the model already does well in its strongest language, it inherits… 38. **[InftyThink: Breaking the Length Limits of Long-Context Reasoning in Large Language Models](https://md.research.elamur.ai/free/en/papers/yan-2025-inftythink-breaking-length-limits-long/)** *(2025 · 29 citations)* Reasoning-oriented LLMs such as DeepSeek-R1 handle hard problems by writing very long chains of thought. This approach has three costs. Attention cost grows quadratically with output length. Generation can be cut off at the maximum context size before an answer appears. Quality also drops once text exceeds the short… 39. **[A Discourse Signal Annotation System for RST Trees](https://md.research.elamur.ai/free/en/papers/gessler-2019-discourse-signal-annotation-system-rst/)** *(2019 · 18 citations)* The paper extends rstWeb, a browser tool for building Rhetorical Structure Theory trees, so that each relation in the tree can carry the signals that let a reader recognise it. 40. **[GumDrop at the DISRPT2019 Shared Task: A Model Stacking Approach to Discourse Unit Segmentation and Connective Detection](https://md.research.elamur.ai/free/en/papers/yu-2019-gumdrop-disrpt2019-shared-task-model/)** *(2019 · 18 citations)* GumDrop is a shared-task system for sentence splitting, elementary discourse unit segmentation and connective detection across 15 corpora in 10 languages (RST, SDRT and PDTB styles). 41. **[AMALGUM – A Free, Balanced, Multilayer English Web Corpus](https://md.research.elamur.ai/free/en/papers/gessler-2020-amalgum-free-balanced-multilayer-english/)** *(2020 · 16 citations)* AMALGUM is an English web corpus of about 4 million tokens, built from openly licensed online sources in eight genres (news, interviews, travel guides, how-to guides, academic papers, fiction, biographies and forum posts), with about 500k tokens per genre. 42. **[Beyond The Wall Street Journal: Anchoring and Comparing Discourse Signals across Genres](https://md.research.elamur.ai/free/en/papers/liu-2019-beyond-wall-street-journal-anchoring/)** *(2019 · 16 citations)* The RST Signalling Corpus records which signals mark each RST relation in news, but not where they are. This pilot anchors signals to tokens in 12 GUM documents (11K tokens) from four genres — academic, how-to, interview, news — with the RST-SC taxonomy extended by visual (image) and textual (title, date… 43. **[Skywork-Math: Data Scaling Laws for Mathematical Reasoning in Large Language Models — The Story Goes On](https://md.research.elamur.ai/free/en/papers/zeng-2024-skywork-math-data-scaling-laws/)** *(2024 · 14 citations)* The paper asks whether strong mathematical reasoning really requires very large models, and whether supervised fine-tuning (SFT) only changes output style. 44. **[Understanding Chain-of-Thought in LLMs through Information Theory](https://md.research.elamur.ai/free/en/papers/ton-2024-understanding-chain-thought-llms-through/)** *(2024 · 14 citations)* Evaluating Chain-of-Thought (CoT) reasoning step by step usually needs costly human labels for every intermediate step, as in process supervision. 45. **[FinanceReasoning: Benchmarking Financial Numerical Reasoning More Credible, Comprehensive and Challenging](https://md.research.elamur.ai/free/en/papers/tang-2025-financereasoning-benchmarking-financial-numerical-reasoning/)** *(2025 · 13 citations)* Benchmarks for financial numerical reasoning have become too easy and too noisy to measure what large reasoning models (LRMs) can actually do. 46. **[Evaluating LLMs' Divergent Thinking Capabilities for Scientific Idea Generation with Minimal Context](https://md.research.elamur.ai/free/en/papers/ruan-2024-evaluating-llms-divergent-thinking-capabilities/)** *(2024 · 13 citations)* Most benchmarks of LLMs on scientific tasks give models rich context, such as full papers or detailed task descriptions. 47. **[PHAnToM: Persona-based Prompting Has An Effect on Theory-of-Mind Reasoning in Large Language Models](https://md.research.elamur.ai/free/en/papers/tan-2024-phantom-persona-based-prompting-has/)** *(2024 · 13 citations)* The paper asks whether role-play prompts that give an LLM a personality change how well it handles Theory-of-Mind (ToM) reasoning, meaning inferring other agents' beliefs and knowledge. 48. **[Scaffolding Coordinates to Promote Vision-Language Coordination in Large Multi-Modal Models](https://md.research.elamur.ai/free/en/papers/lei-2025-scaffolding-coordinates-promote-vision-language/)** *(2025 · 11 citations)* Large multi-modal models such as GPT-4V still struggle with tasks where they must relate several pieces of visual information and express that reasoning accurately in text. 49. **[Skywork R1V2: Multimodal Hybrid Reinforcement Learning for Reasoning](https://md.research.elamur.ai/free/en/papers/wang-2025-skywork-r1v2-multimodal-hybrid/)** *(2025 · 10 citations)* "Slow-thinking" vision-language models tend to improve on math and science benchmarks, but they often lose accuracy on everyday perception tasks and hallucinate more about image content. 50. **[SPPO: Sequence-Level PPO for Long-Horizon Reasoning Tasks](https://md.research.elamur.ai/free/en/papers/wang-2026-sppo-sequence-level-ppo-long/)** *(2026 · 8 citations)* Reinforcement learning with verifiable rewards is a common way to improve LLM reasoning, but the two main options each have a cost. [← prev](https://md.research.elamur.ai/free/en/authors/yang-liu/) [1](https://md.research.elamur.ai/free/en/authors/yang-liu/) **2** [3](https://md.research.elamur.ai/free/en/authors/yang-liu/page/3/) [4](https://md.research.elamur.ai/free/en/authors/yang-liu/page/4/) [next →](https://md.research.elamur.ai/free/en/authors/yang-liu/page/3/)