Yang Liu · page 3
Papers
- LogicPro: Improving Complex Logical Reasoning via Program-Guided Learning (2025 · 8 citations)
Large language models often fail at multi-step logical reasoning when a task is posed in plain text, even when the same task is easy to solve with code. - CORBA: Contagious Recursive Blocking Attacks on Multi-Agent Systems Based on Large Language Models (2026 · 7 citations)
The paper identifies a security gap in LLM-based multi-agent systems (LLM-MASs). Prior work mostly studied harmful content or attacks on single models and services. Here the authors target the cooperation between agents. They define a new threat class, Denial-of-Collaboration (DoC): the system keeps running but does… - A Neural Approach to Discourse Relation Signal Detection (2020 · 7 citations)
Discourse relations such as CONDITION, CONCESSION or CAUSE are often marked by connectives. Corpus-wide frequency or PMI counts give each word type a single score. They cannot tell whether a given occurrence of "and" or "to" actually carries a relation, and they cannot flag words that point the wrong way. Zeldes and… - Measuring Language Development in Early Childhood Education: A Case Study of Grammar Checking in Child Language Transcripts (2011 · 7 citations)
Common measures of preschoolers' language development, such as IPSyn, take into account only correctly used grammatical constructions. - G2: Guided Generation for Enhanced Output Diversity in LLMs (2025 · 6 citations)
LLMs often return near-identical answers when asked the same question repeatedly. This hurts creative writing, best-of-N selection, synthetic data creation and test-time scaling for reasoning. Raising the sampling temperature adds variety but lowers quality, and re-prompting with earlier answers can drift away from… - Skywork-R1V3 Technical Report (2025 · 6 citations)
Open-source vision-language models still trail proprietary systems on complex visual reasoning, partly because high-quality multimodal annotations are scarce. - Writing-RL: Advancing Long-form Writing via Adaptive Curriculum Reinforcement Learning (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… - Automatically Detecting Reduced-formed English Pronunciations by Using Deep Learning (2022 · 4 citations)
The paper addresses automatic detection of reduced-form pronunciations in the speech of Chinese learners of English. - Beyond “I Don't Know”: Evaluating LLM Self-Awareness in Discriminating Data and Model Uncertainty (2026 · 3 citations)
Most work on LLM abstention treats every refusal the same way. This paper argues that a reliable model also needs to know why it is unsure. It separates two sources of uncertainty. Data uncertainty means the question is ambiguous or missing information, which calls for asking the user to clarify. Model uncertainty… - Skywork-SWE: Unveiling Data Scaling Laws for Software Engineering in LLMs (2025 · 3 citations)
Training LLM agents to fix real GitHub issues is held back by scarce data. Every task needs an executable environment and validated unit tests, and building these by hand is slow, so most public SWE datasets have only a few thousand instances. The authors created an automated, incremental pipeline with three stages… - Guided by the Plan: Enhancing Faithful Autoregressive Text-to-Audio Generation with Guided Decoding (2026 · 2 citations)
Autoregressive (AR) text-to-audio generators produce temporally coherent sound. However, they often miss or misplace parts of complex prompts that describe several sound events. Diffusion models handle this better because they see the whole timeline at once. The authors probe the AR model Siren and find that its… - Attention-based sequence-to-sequence model for speech recognition: development of state-of-the-art system on LibriSpeech and its application to non-native English (2018 · 2 citations)
The authors ask whether an end-to-end attention model can match or beat classic hybrid HMM/NN recognizers, especially for English spoken by learners. - MUSEG: Reinforcing Video Temporal Understanding via Timestamp-Aware Multi-Segment Grounding (2026 · 1 citation)
Multimodal LLMs remain weak at fine-grained temporal reasoning in video, such as locating when events happen. - Scaling External Knowledge Input Beyond Context Windows of LLMs via Multi-Agent Collaboration (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. - Streaming Hallucination Detection in Long Chain-of-Thought Reasoning (2026 · 1 citation)
Errors in long chain-of-thought (CoT) reasoning tend to appear gradually and spread across steps, often hidden behind locally convincing text. - Overview of AMALGUM – Large Silver Quality Annotations across English Genres (2021 · 1 citation)
Discourse-oriented research faces a gap. Huge web corpora carry only shallow annotation and have unclear makeup, while richly annotated corpora tend to be small. AMALGUM (A Machine-Annotated Lookalike of GUM) is meant to sit between the two. The authors used the gold GUM corpus (130K tokens, eight genres) as a… - Enabling Stroke-Level Structural Analysis of Hieroglyphic Scripts without Language-Specific Priors (2026)
Logographic and hieroglyphic scripts carry meaning in how strokes are arranged inside each character. Current LLMs treat characters as opaque tokens, and MLLMs see them only as pixel grids, so neither captures stroke structure. Earlier structural methods are usually tied to one script and depend on costly annotation… - Enhancing Lexical Relation Mining with Structured Sememe Knowledge (2026)
The paper tackles Lexical Relation Mining (LRM) in two settings. Lexical Relation Classification (LRC) labels word pairs with relations such as synonymy or hypernymy. Lexical Entailment (LE) scores how strongly one word entails another. Current PLM-based systems still mix up subtle relations, for example synonymy… - FlowHN: Adaptive Token Routing for Efficient Parallel Hybrid Networks (2026)
Hybrid language models pair attention, which handles precise token interactions but scales quadratically, with state-space models (SSMs), which scale linearly but are weaker at fine-grained retrieval. - GIFT: Guided Fine-Tuning and Transfer for Enhancing Instruction-Tuned Language Models (2026)
Adapting instruction-tuned LLMs to a specific domain is risky: fine-tuning them directly can disturb the balance of skills built during post-training. - InstructDiff: Domain-Adaptive Data Selection via Contrastive Entropy for Efficient LLM Fine-Tuning (2026)
Fine-tuning LLMs on entire instruction datasets is expensive and yields diminishing returns. Existing data-selection techniques also tend to work only in the domain they were designed for: methods tuned for general instruction-following transfer poorly to reasoning, and vice versa. InstructDiff addresses this with a… - MASFactory: A Graph-centric Framework for Orchestrating LLM-Based Multi-Agent Systems with Vibe Graphing (2026)
Building LLM-based multi-agent systems (MAS) as directed graphs in tools like LangGraph or Dify still takes a lot of manual work. - Observations and Remedies for Large Language Model Bias in Self-Consuming Performative Loop (2026)
LLMs are increasingly trained on text written by earlier models. In deployment, how well a model serves each user group also affects who keeps sending queries. The authors formalize this as the Self-Consuming Performative Loop (SCPL). Each new model generation learns from synthetic responses to prompts, and the group… - Revisiting a Pain in the Neck: A Semantic Reasoning Benchmark for Language Models (2026)
SemanticQA is an English benchmark that tests how well language models handle semantic phrases, also called multiword expressions (MWE). - Trustworthy and Explainable Causal Representation Learning in Transformers (2026)
Explanations for Transformers are often built by learning an importance mask over input features and applying it to the input representation.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/yang-liu/page/3/