Papers by Yifan Zheng

31 papers
Graph Reasoning for Question Answering with Triplet Retrieval (2023.findings-acl)

Copied to clipboard

Challenge: Existing methods to answer complex questions require reasoning over knowledge graphs (KGs) state-of-the-art methods constrain retrieved knowledge in local subgraphs and discard more diverse triplets that are disconnected but useful for question answering.
Approach: They propose a method to retrieve the most relevant triplets from KGs and then rerank them, which are then concatenated with questions to be fed into language models.
Outcome: The proposed method outperforms state-of-the-art methods on commonsenseQA and OpenbookQA datasets with 4.6% absolute accuracy.
AdaZeta: Adaptive Zeroth-Order Tensor-Train Adaption for Memory-Efficient Large Language Models Fine-Tuning (2024.emnlp-main)

Copied to clipboard

Challenge: Recent advances in memory-efficient zeroth-order methods have limited their widespread adoption due to performance drops and a high risk of divergence.
Approach: They propose a memory-efficient zeroth-order framework to improve performance and convergence of the MeZO methods by using only forward passes.
Outcome: The proposed framework improves performance and convergence of the proposed methods on Roberta-Large and Llama-2-7B models.
Wanda++: Pruning Large Language Models via Regional Gradients (2025.findings-acl)

Copied to clipboard

Challenge: Existing pruning methods suffer from accuracy degradation without full-model sparsity-aware fine-tuning.
Approach: They propose a pruning framework that uses decoder-block-level regional gradients to improve pruning accuracy.
Outcome: The proposed pruning framework outperforms the state-of-the-art pruning frameworks by utilizing decoder-block-level regional gradients.
Aligning Large Language Models with Implicit Preferences from User-Generated Content (2025.acl-long)

Copied to clipboard

Challenge: Existing preference learning methods rely heavily on curated data from humans or advanced LLMs, which is costly and difficult to scale.
Approach: They propose a framework that leverages implicit preferences in unlabeled user-generated content to generate preference data.
Outcome: The proposed framework transforms user-generated content into user queries and generates responses from the policy model.
Hephaestus: Improving Fundamental Agent Capabilities of Large Language Models through Continual Pre-Training (2025.naacl-long)

Copied to clipboard

Challenge: Existing LLMs often rely on complex prompting or extensive fine-tuning to introduce new capabilities while preserving strong generalizability.
Approach: They propose a large-scale pre-training corpus to enhance LLM agents' capabilities . they use 103B agent-specific data encompassing 76,537 APIs .
Outcome: The proposed training corpus outperforms open-source LLMs and commercial LLM agents on three agent benchmarks.
Knowledge-Selective Pretraining for Attribute Value Extraction (2023.findings-emnlp)

Copied to clipboard

Challenge: Existing methods for AVE are limited on rare attributes due to poor generalization ability.
Approach: They propose to leverage pretraining and transfer learning to address weaknesses in existing methods.
Outcome: The proposed method achieves new state-of-the-art performance without pretraining on rare attributes with limited training resources.
QuZO: Quantized Zeroth-Order Fine-Tuning for Large Language Models (2025.emnlp-main)

Copied to clipboard

Challenge: Large Language Models (LLMs) are quantized to lower precision to reduce memory cost and latency in inference.
Approach: They propose a quantized zeroth-order framework for fine-tuning Large Language Models (LLMs) using low-precision forward passes.
Outcome: The proposed method achieves comparable results to first-order methods in FP8 and superior accuracy in INT8 and INT4 training.
IHEval: Evaluating Language Models on Following the Instruction Hierarchy (2025.naacl-long)

Copied to clipboard

Challenge: Instruction-tuned language models (LMs) are increasingly deployed as interactive services across various applications.
Approach: They propose a benchmark to evaluate models' ability to follow the instruction hierarchy by comparing their models to a set of benchmarks.
Outcome: The proposed benchmark covers 3,538 examples across nine tasks covering cases where instructions in different priorities either align or conflict.
UniConv: Unifying Retrieval and Response Generation for Large Language Models in Conversations (2025.acl-long)

Copied to clipboard

Challenge: Existing conversational search systems are usually built with two different models . this separation restricts the system from leveraging the model's intrinsic knowledge simultaneously . Existing studies for developing unified models cannot fully address the aspects of understanding conversational context, managing retrieval independently, and generating responses.
Approach: They propose to unify dense retrieval and response generation for large language models in conversation by fine-tuning and mitigating data discrepancy.
Outcome: The proposed model can outperform existing models on five conversational search datasets and reduce inconsistency risks while mitigating data discrepancy.
MemeReaCon: Probing Contextual Meme Understanding in Large Vision-Language Models (2025.emnlp-main)

Copied to clipboard

Challenge: Current approaches focus on isolated meme analysis, either for harmful content detection or standalone interpretation, overlooking a fundamental challenge: the same meme can express different intents depending on its conversational context.
Approach: They propose a benchmark to evaluate how large vision language models understand memes in their original context.
Outcome: The proposed benchmark evaluates how large vision language models understand meme intent in their original context.
FLAT-LLM: Fine-grained Low-rank Activation Space Transformation for Large Language Model Compression (2026.findings-eacl)

Copied to clipboard

Challenge: Low-rank decomposition methods suffer from accuracy degradation and expensive calibration procedures.
Approach: They propose a fast and accurate, training-free structural compression method based on fine-grained low-rank transformations in the activation space.
Outcome: The proposed method outperforms pruning baselines in generalization and downstream performance while delivering inference speedups.
Can Language Models Follow Multiple Turns of Entangled Instructions? (2025.findings-emnlp)

Copied to clipboard

Challenge: Despite of significant achievements in improving instruction-following capabilities of large language models, the ability to process multiple potentially entangled or conflicting instructions remains a considerable challenge.
Approach: They construct multi-turn instruction with 1.1K high-quality multi-turned conversations using the human-in-the-loop approach and examine their capabilities.
Outcome: The proposed model shows that it is difficult to integrate multiple turns and balance competing objectives when instructions intersect or conflict.
PRGC: Potential Relation and Global Correspondence Based Joint Relational Triple Extraction (2021.acl-long)

Copied to clipboard

Challenge: Recent methods for extracting entities and relations from unstructured texts suffer from limitations, such as redundancy of relation prediction and inefficiency.
Approach: They propose a joint relational triple extraction framework based on Potential Relation and Global Correspondence (PRGC) they propose overlapping triples for relation prediction and relation-relational alignment .
Outcome: The proposed framework achieves state-of-the-art performance on public benchmarks with higher efficiency and consistent performance gain on complex scenarios of overlapping triples.
Improving Consistency for Text Summarization with Energy Functions (2023.findings-emnlp)

Copied to clipboard

Challenge: Current abstractive summarization models generate inconsistent content due to the inherently noisy dataset and the discrepancy between maximum likelihood estimation based training objectives and consistency measurements.
Approach: They propose a new consistency taxonomy that categorizes inconsistent content into faithfulness, factuality, and self-supportiveness.
Outcome: Experiments on XSUM and CNN/DM datasets show that EnergySum mitigates the trade-off between accuracy and consistency.
Rethinking Token Reduction for State Space Models (2024.emnlp-main)

Copied to clipboard

Challenge: Existing methods for token reduction for SSMs lead to performance drops . a recent study shows that Mamba-2 improves the accuracy of the model by 5.7% to 13.1% .
Approach: They propose a token reduction method that integrates token importance and similarity into SSMs and takes advantage of pruning and merging.
Outcome: The proposed method improves accuracy by 5.7% to 13.1% on six benchmarks with Mamba-2 compared to existing methods while reducing computational demands and memory requirements.
Sentient Agent as a Judge: Evaluating Higher-Order Social Cognition in Large Language Models (2026.findings-acl)

Copied to clipboard

Challenge: Large language models (LLMs) have evolved from statistical sequence predictors to sophisticated autonomous agents capable of reasoning, planning, and sustaining multi-turn conversa-tions.
Approach: They propose a system that instantiates a "Sentient Agent" that simulates human-like emotional changes and inner thoughts to provide a more realistic evaluation of the model in multi-turn conversations.
Outcome: The proposed framework measures the agent's higher-order social cognition in multi-turn conversations.
Controlling Styles in Neural Machine Translation with Activation Prompt (2023.findings-acl)

Copied to clipboard

Challenge: Earlier studies on controlling styles in neural machine translation (NMT) have focused on regulating the level of formality, but they still encounter two major challenges.
Approach: They propose a method to control the style of neural machine translation by retrieving prompts from stylized monolingual corpus.
Outcome: The proposed method can control the style of translation and achieve remarkable performance.
MuBench: Assessment of Multilingual Capabilities of Large Language Models Across 61 Languages (2026.findings-acl)

Copied to clipboard

Challenge: Existing evaluation datasets lack cross-lingual alignment, leaving assessments of multilingual capabilities fragmented in both language and skill coverage.
Approach: They propose to use multilingual consistency as a complementary metric to assess performance bottlenecks and guide model improvement.
Outcome: The proposed model lacks cross-lingual alignment and language coverage gaps between state-of-the-art models.
Large Language Models Are Poor Clinical Decision-Makers: A Comprehensive Benchmark (2024.emnlp-main)

Copied to clipboard

Challenge: Existing studies focus on evaluating large language models in close-ended QA tasks, but many clinical decisions involve answering open-ended questions without pre-set options.
Approach: They construct a benchmark to better understand large language models in the clinic . they use existing datasets to evaluate LLMs in clinical situations .
Outcome: The proposed model outperforms human experts in multiple medical tasks.
BFS-Prover: Scalable Best-First Tree Search for LLM-based Automatic Theorem Proving (2025.acl-long)

Copied to clipboard

Challenge: Existing approaches to theorem proving in large language models rely on value functions and/or Monte Carlo Tree Search (MCTS), but the potential of simpler methods like Best-First Tree Search remains underexplored.
Approach: They propose a scalable expert iteration framework that implements strategic data filtering at each expert iteration round, excluding problems solvable via beam search node expansion to focus on harder cases.
Outcome: The proposed framework achieves a state-of-the-art score of 72.95 on the MiniF2F test set and challenges the perceived necessity of complex tree search methods.
Gradient-Attention Guided Dual-Masking Synergetic Framework for Robust Text-based Person Retrieval (2025.emnlp-main)

Copied to clipboard

Challenge: a large-scale visionlanguage pre-training framework is limited by the scarcity of large-sized annotated vision-language data . noise-resistant data construction pipeline is needed to filter and caption web-sourced images . noisy text tokens can be a problem for fine-grained representation learning .
Approach: They develop a noise-resistant data construction pipeline that leverages in-context learning capabilities of MLLMs to automatically filter and caption web-sourced images.
Outcome: The proposed framework improves cross-modal alignment by masking noisy textual tokens based on the gradient-attention similarity score.
Communication-Efficient and Tensorized Federated Fine-Tuning of Large Language Models (2025.findings-acl)

Copied to clipboard

Challenge: Large Language Models (LLMs) excel in translation and summarization due to the capabilities of transformer architectures.
Approach: They propose to integrate tensorized adapters into model encoder/decoder blocks to improve model adaptability against data heterogeneity.
Outcome: Experiments on large-scale cross-device FL and large-silo FL show that the proposed methods perform on par or even better than existing federated PEFT approaches while reducing communication cost.
Render-of-Thought: Rendering Textual Chain-of-Thought as Images for Visual Latent Reasoning (2026.acl-long)

Copied to clipboard

Challenge: Recent work on Chain-of-Thought prompting imposes substantial computational overhead . lack of supervision obscures the analyzability of the latent reasoning chain.
Approach: They propose a framework to render latent reasoning chain into images, making latent rationale explicit and traceable.
Outcome: The proposed framework achieves 3-4 token compression and substantial inference acceleration compared to explicit CoT prompting.
MaZO: Masked Zeroth-Order Optimization for Multi-Task Fine-Tuning of Large Language Models (2025.emnlp-main)

Copied to clipboard

Challenge: Large language models (LLMs) have demonstrated exceptional capabilities across diverse tasks, but their fine-tuning requires significant memory, posing challenges for resource-constrained environments.
Approach: They propose a ZO-based framework that eliminates the need for backpropagation and provides a memory-efficient alternative to backprograming.
Outcome: The proposed framework surpasses first-order methods in performance and accuracy.
SCOTT: Self-Consistent Chain-of-Thought Distillation (2023.acl-long)

Copied to clipboard

Challenge: Large language models (LMs) generate free-text rationales for their predictions via chain-of-thought prompting, but there is little guarantee that the generated rationale is consistent with LM’s predictions or faithfully justify the decisions.
Approach: They propose a faithful knowledge distillation method to learn a small, self-consistent CoT model from a larger teacher model by contrastive decoding.
Outcome: The proposed method yields comparable performance but is less faithful than baselines.
LoRETTA: Low-Rank Economic Tensor-Train Adaptation for Ultra-Low-Parameter Fine-Tuning of Large Language Models (2024.naacl-long)

Copied to clipboard

Challenge: Existing methods for parameter-efficient fine-tuning are limited by the growing number of trainable parameters with the rapid deployment of Large Language Models (LLMs).
Approach: They propose a parameter-efficient framework that reduces trainable parameters through tensor-train decomposition.
Outcome: The proposed methods achieve comparable or better performance than most widely used methods with up to 100 fewer parameters on the LLaMA-2-7B models.
SessionIntentBench: A Multi-task Inter-session Intention-shift Modeling Benchmark for E-commerce Customer Behavior Understanding (2026.findings-acl)

Copied to clipboard

Challenge: Existing models fail to capture and model customer intention effectively because of insufficient information exploitation and only apparent information like descriptions and titles are used.
Approach: They propose to exploit existing session data to capture and model intention in E-commerce product purchase sessions using a multimodal benchmark.
Outcome: The proposed framework can bridge the gap between intention understanding in simplified research cases like co-buy intention and more complex yet practical scenarios like session history.
FolkScope: Intention Knowledge Graph Construction for E-commerce Commonsense Discovery (2023.findings-acl)

Copied to clipboard

Challenge: Existing intention-based studies on recommendation tasks are limited and use models to implicitly model the intention memberships.
Approach: They propose a framework that leverages the generation power of large language models and human-in-the-loop annotation to semi-automatically construct the intention knowledge graph.
Outcome: The proposed framework can model e-commerce knowledge and have many potential applications.
Supporting Medical Relation Extraction via Causality-Pruned Semantic Dependency Forest (2022.coling-1)

Copied to clipboard

Challenge: Medical relation extraction (MRE) tasks aims to extract relations between entities in medical literature.
Approach: They propose to combine semantic and syntactic information from medical texts by using causal explanation theory.
Outcome: Empirically, the proposed model outperforms existing methods on benchmark medical datasets.
Retrieval-Augmented Multilingual Keyphrase Generation with Retriever-Generator Iterative Training (2022.findings-naacl)

Copied to clipboard

Challenge: Existing studies on keyphrase generation on non-English languages haven’t been vastly investigated.
Approach: They propose a retrieval-augmented method for multilingual keyphrase generation that leverages keyphrase annotations in English datasets to facilitate generating keyphrases in low-resource languages.
Outcome: The proposed model outperforms baselines on non-English keyphrase generation datasets and the proposed model is scalable.
QUARTZ: Quantile-Aware Routing and Queueing for TTFT SLOs in LLM Serving (2026.findings-acl)

Copied to clipboard

Challenge: Prefill costs scale with prompt length and decode lengths are uncertain, and prefix locality creates strong performance skew across requests.
Approach: They propose a quantile-aware routing and queueing layer that predicts conservative quantiles rather than point estimates using lightweight router-visible signals.
Outcome: The proposed layer predicts conservative quantile-based request-cost proxies, rather than point estimates, using lightweight router-visible signals.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations