Zhiyuan Liu · page 2
Papers
- Prompt-learning for Fine-grained Entity Typing (2022 · 181 citations)
The paper asks whether prompt-learning, which recasts a downstream task as a cloze-style masked-word prediction, can replace classifier-based fine-tuning for fine-grained entity typing. It focuses on cases with little or no labelled data. The authors build a simple pipeline called PLET with two parts: - CLEVE: Contrastive Pre-training for Event Extraction (2021 · 155 citations)
Event extraction (EE) means locating event triggers, assigning event types, and identifying arguments with their roles. - Continual Relation Learning via Episodic Memory Activation and Reconsolidation (2020 · 152 citations)
Relation extraction models are usually trained once on a fixed set of relation types, but new relations keep appearing in real text. - ERICA: Improving Entity and Relation Understanding for Pre-trained Language Models via Contrastive Learning (2021 · 139 citations)
Standard pre-training objectives such as masked language modeling treat words as the basic unit. They do not explicitly teach models how entities relate to each other across a document. Citing DocRED-style statistics, the authors note that over 40% of relational facts can only be recovered by combining several… - InfLLM: Training-Free Long-Context Extrapolation for LLMs with an Efficient Context Memory (2024 · 126 citations)
LLMs pre-trained on windows of only a few thousand tokens break down on much longer inputs. Unseen positions push them out of distribution, and large amounts of irrelevant context distract attention. The usual fix, continued pre-training on long sequences, is expensive and can hurt short-context skills. InfLLM avoids… - MoEfication: Transformer Feed-forward Layers are Mixtures of Experts (2022 · 115 citations)
Feed-forward networks (FFNs) hold roughly two-thirds of a Transformer's parameters, yet how they compute has received far less study than attention. - MAVEN-ERE: A Unified Large-scale Dataset for Event Coreference, Temporal, Causal, and Subevent Relation Extraction (2022 · 96 citations)
Event relation extraction (ERE) covers coreference, temporal, causal and subevent links between events. Progress has been limited by datasets that are small (hundreds of documents) and that label only one or two relation types at a time. This prevents models from exploiting how relations constrain each other; for… - Does William Shakespeare REALLY Write Hamlet? Knowledge Representation Learning with Confidence (2017 · 93 citations)
Large knowledge graphs such as Freebase are increasingly filled by automatic extraction and crowdsourcing, so they contain wrong or conflicting facts. - Finding Skill Neurons in Pre-trained Transformer-based Language Models (2022 · 85 citations)
Pre-trained Transformers perform well across many NLP tasks, but it is unclear where task-relevant abilities sit among their parameters. - Simulating Classroom Education with LLM-Empowered Agents (2025 · 84 citations)
SimClass is a framework that tests whether a group of LLM-driven agents can run a convincing virtual classroom with real learners. - CREATOR: Tool Creation for Disentangling Abstract and Concrete Reasoning of Large Language Models (2023 · 75 citations)
Tool-augmented LLMs usually depend on a fixed set of existing APIs. They must plan and execute in a single pass of implicit reasoning, which breaks easily, and they handle errors poorly. CREATOR addresses this by having the model write its own tools instead of only calling given ones. The framework has four stages.… - RAGEval: Scenario Specific RAG Evaluation Dataset Generation Framework (2025 · 34 citations)
Testing RAG systems in specialised fields such as law, finance or medicine is hard. Building expert-annotated, domain-specific test sets is expensive and raises privacy issues, and lexical metrics like BLEU or ROUGE-L judge long answers poorly. RAGEval addresses this by generating evaluation data automatically.… - Sub-Character Tokenization for Chinese Pretrained Language Models (2023 · 29 citations)
Chinese pretrained language models usually tokenize text either one character at a time or by running sub-word algorithms directly on raw characters. - ChartCoder: Advancing Multimodal Large Language Model for Chart-to-Code Generation (2025 · 24 citations)
Open-source multimodal LLMs read charts reasonably well, but they struggle to turn a chart image into plotting code. Their output often fails to run, picks the wrong chart type, or loses details such as axis sizes. The authors argue that code is a lossless description of a chart, unlike textual summaries. - Search and Refine During Think: Facilitating Knowledge Refinement for Improved Retrieval-Augmented Reasoning (2025 · 18 citations)
Retrieval-augmented LLMs trained with reinforcement learning tend to reason over whole retrieved documents, which are often noisy. - ConPET: Continual Parameter-Efficient Tuning for Large Language Models (2023 · 14 citations)
Adapting large language models to a stream of new tasks is costly when every update changes all weights. Full fine-tuning also tends to erase skills learned on earlier tasks, a problem known as catastrophic forgetting. Song et al. propose ConPET, a framework for continual task adaptation of LLMs in which training… - ThinkNote: Enhancing Knowledge Integration and Utilization of Large Language Models via Constructivist Cognition Modeling (2026 · 2 citations)
LLMs often struggle to use retrieved external knowledge reliably. Noisy, incomplete or unfamiliar evidence can produce inconsistent answers and hallucinations. ThinkNote addresses this by treating the model as an active builder of knowledge rather than a passive reader of context, drawing on constructivist learning… - Enhancing Long-Chain Reasoning Distillation through Error-Aware Self-Reflection (2026 · 1 citation)
Distilling long Chain-of-Thought (CoT) traces from strong reasoning models into small models often works poorly. - StateX: Enhancing RNN Recall via Post-training State Expansion (2026 · 1 citation)
Recurrent language models such as linear attention and state-space models process long inputs at constant cost per token. - UltraEval-Audio: A Unified Framework for Comprehensive Evaluation of Audio Foundation Models (2026 · 1 citation)
Evaluating audio foundation models (AFMs) is difficult because benchmarks and evaluation code are scattered across sources.
This page in Markdown → https://md.research.elamur.ai/free/en/authors/zhiyuan-liu/page/2/