Papers by Yan Fang

24 papers
Thinking with DistilQwen: A Tale of Four Distilled Reasoning and Reward Model Series (2025.emnlp-industry)

Copied to clipboard

Challenge: In the rapidly evolving landscape of large language models, the need for efficient reasoning models has become increasingly urgent.
Approach: They extend the Qwen model family by introducing four model series specifically designed for industrial applications.
Outcome: The proposed models outperform previous models in multiple benchmarks and provide scalable training and inference functionality on the Alibaba Cloud PAI platform.
CART: A Generative Cross-Modal Retrieval Framework With Coarse-To-Fine Semantic Modeling (2025.acl-long)

Copied to clipboard

Challenge: Cross-modal retrieval tasks are used to retrieve data from one modality or another based on a query from another modality.
Approach: They propose a generative cross-modal retrieval framework based on coarse-to-fine semantic modeling . they propose combining K-Means and RQ-VAE to discretize multimodal data into token sequences that support autoregressive generation.
Outcome: The proposed framework achieves excellent performance and efficiency in multimodal retrieval tasks.
Model Composition for Multimodal Large Language Models (2024.acl-long)

Copied to clipboard

Challenge: Existing methods for creating versatile MLLMs rely on joint training with paired instruction data, which is resource-intensive and challenging to extend to new modalities.
Approach: They propose a new paradigm for multimodal large language models by reusing modality encoders and merging LLM parameters.
Outcome: The proposed model retains the modal understanding capabilities of each original model.
TrInk: Ink Generation with Transformer Network (2025.emnlp-main)

Copied to clipboard

Challenge: Existing methods for handwriting generation capture global dependencies and can generate high-quality handwritten samples.
Approach: They propose a Transformer-based model for ink generation, TrInk, which captures global dependencies.
Outcome: The proposed model reduces character error rate and word error rate by 35.56% on the IAM-OnDB dataset compared to previous models.
LLaMA-Omni 2: LLM-based Real-time Spoken Chatbot with Autoregressive Streaming Speech Synthesis (2025.acl-long)

Copied to clipboard

Challenge: LLaMA-Omni 2 is a series of speech language models (SpeechLMs) based on large language models.
Approach: They introduce a series of speech language models capable of real-time speech interaction . LLaMA-Omni 2 trains on 200K multi-turn speech dialogue samples .
Outcome: The proposed speech language models surpass state-of-the-art models on spoken question answering and speech instruction.
Revealing and Enhancing Core Visual Regions: Harnessing Internal Attention Dynamics for Hallucination Mitigation in LVLMs (2026.findings-acl)

Copied to clipboard

Challenge: Existing training-free methods are vulnerable to the attention sink phenomenon . Existing methods include contrastive decoding and auxiliary expert models .
Approach: They propose a training-free attention intervention that constructs a PAD map to identify semantically core visual regions and applies per-head Median Absolute Deviation Scaling to adaptively control the intervention strength.
Outcome: The proposed intervention improves visual grounding and reduces hallucinations on multiple LVLMs and benchmarks.
Enhancing Reasoning Abilities of Small LLMs with Cognitive Alignment (2025.emnlp-main)

Copied to clipboard

Challenge: Existing methods to distill chain-of-thought (CoT) results from large language reasoning models (LRMs) to small models are ineffective and require substantial amount of annotated data.
Approach: They propose a Critique-Rethink-Verify system for training small language reasoning models that can be critiquized according to the cognitive capabilities of smaller models.
Outcome: The proposed system outperforms other methods on challenging reasoning benchmarks.
LEGO-ABSA: A Prompt-based Task Assemblable Unified Generative Framework for Multi-task Aspect-based Sentiment Analysis (2022.coling-1)

Copied to clipboard

Challenge: Existing generative methods focus on a single task at a time.
Approach: They propose a unified generative multi-task framework that can solve multiple ABSA tasks . they propose to control the type of task prompts consisting of multiple element prompts .
Outcome: The proposed framework achieves state-of-the-art results in almost all ABSA tasks and competitive results in task transfer scenarios.
CTC-based Non-autoregressive Textless Speech-to-Speech Translation (2024.findings-acl)

Copied to clipboard

Challenge: Existing direct speech-to-speech translation models require text supervision during training, which is not feasible for numerous unwritten languages.
Approach: They propose a non-autoregressive (NAR) model that generates discrete units from the source speech and employs a unit-based vocoder to synthesize the target.
Outcome: The proposed model achieves translation quality comparable to the autoregressive model while preserving up to 26.81 decoding speedup.
Efficient Training for Cross-lingual Speech Language Models (2026.findings-acl)

Copied to clipboard

Challenge: Currently, large language models (LLMs) focus on the text modality, making speech modeling difficult.
Approach: They propose a cross-lingual speech language model that trains on discrete speech tokens to achieve cross-modal and cross-linguistic alignment through continual pre-training.
Outcome: The proposed method achieves cross-modal and cross-lingual alignment through continual pre-training.
Listening Like Humans: Semantics-Guided Noise-Robust Multimodal Speech Recognition (2026.acl-long)

Copied to clipboard

Challenge: Severe acoustic degradation results in unreliable ASR outputs . et al., 2024b): critical concerns regarding reliability and fairness of ASR .
Approach: They propose a multimodal framework that reframes ASR as semantics-guided speech reconstruction.
Outcome: The proposed framework achieves an average reduction in WER while also attaining 98.71% BERTScore and 96.7% USE over advanced baselines.
CESRec: Constructing Pseudo Interactions for Sequential Recommendation via Conversational Feedback (2025.findings-emnlp)

Copied to clipboard

Challenge: Existing Sequential Recommendation Systems (SRS) rely on collaborative filtering signals and fail to capture real-time user preferences.
Approach: They propose a framework that integrates the long-term preference modeling of SRS with the real-time preference elicitation of CRS.
Outcome: The proposed framework integrates the long-term preference modeling of SRS with the real-time preference elicitation of CRS.
Injecting Domain-Specific Knowledge into Large Language Models: A Comprehensive Survey (2025.findings-emnlp)

Copied to clipboard

Challenge: specialized LLMs are often limited in domain-specific applications that require specialized knowledge.
Approach: They provide a comprehensive overview of four key methods to enhance large language models by integrating domain-specific knowledge.
Outcome: The proposed methods are categorized into four key approaches: dynamic knowledge injection, static knowledge embedding, modular adapters, and prompt optimization.
Reinforcement Learning on Pre-Training Data (2026.acl-long)

Copied to clipboard

Challenge: Recent progress in large language models is driven by scaling of training compute through pre-training with nexttoken prediction (NTP) or post-training (RL) Pre-training using NTP enables models to acquire extensive knowledge and skills from general data, but it suffers from data inefficiency and catastrophic forgetting in continual learning settings.
Approach: They propose to scale training compute through pre-training with next-token prediction (NTP) or post-training by scaling reinforcement learning (RL) to improve learning from general data.
Outcome: Experiments on multiple benchmarks and models show that the proposed approach improves continual pre-training and provides a strong foundation for post-training on Qwen3-8B-Base.
EduMARS: Can Vision-Language Models Grade Like Teachers? Benchmarking Multimodal, Rubric-Based Assessment on Chinese K-12 Answers (2026.findings-acl)

Copied to clipboard

Challenge: Existing benchmarks for automated grading of student work fail to evaluate real student responses . existing models fail to assess real student work, especially on cognitively demanding tasks .
Approach: They propose a multimodal benchmark for rubric-aligned evaluation of real Chinese K-12 student answers.
Outcome: The proposed model improves performance and interpretability of existing models on EduMARS . existing models fail to perform on real-world, cognitively demanding tasks, authors say .
Enhancing Multimodal Unified Representations for Cross Modal Generalization (2025.findings-acl)

Copied to clipboard

Challenge: Existing studies on discrete unified representations overlook important distinctions between different dimensions of features.
Approach: They propose to use a codebook to optimize unified representations from pretraining and fine- and coarse-grained disentangling to optimize the representations.
Outcome: The proposed methods improve the interpretability of multimodal unified representations . they use training-free optimization of codebook and fine and coarse cross-modal disentangling .
AscendKernelGen: LLM-Driven Kernel Generation for NPUs (2026.findings-acl)

Copied to clipboard

Challenge: Neural Processing Units (NPUs) are critical for AI infrastructure, but their development remains a bottleneck due to vendor-specific Domain-Specific Languages (DSLs).
Approach: They propose a framework for NPU kernel development that bridges the gap in hardware-specific coding . compiler success on complex Level-2 kernels improves from 0% to 95.5%, they say .
Outcome: The proposed framework bridges the gap in hardware-specific coding, showing a near-zero success rate on complex kernels.
Vision-Language Introspection: Mitigating Overconfident Hallucinations in MLLMs via Interpretable Bi-Causal Steering (2026.acl-long)

Copied to clipboard

Challenge: Existing approaches to overcome object hallucination are limited . Existing mitigations include costly retraining and a training-free inference framework .
Approach: They propose a training-free inference framework that simulates a metacognitive self-correction process.
Outcome: The proposed framework reduces object hallucination rates by 12.67% on MMHal-Bench and improves accuracy by 5.8% on POPE.
AMPO: Automatic Multi-Branched Prompt Optimization (2024.emnlp-main)

Copied to clipboard

Challenge: Existing prompt engineering techniques are limited to producing single flow instructions, struggling with handling diverse patterns.
Approach: They propose an automatic prompt optimization method that iteratively develops a multi-branched prompt using failure cases as feedback.
Outcome: The proposed method achieves the best results across five tasks and demonstrates significant optimization efficiency due to adoption of a minimal search strategy.
Attention Entropy is a Key Factor: An Analysis of Parallel Context Encoding with Full-attention-based Pre-trained Language Models (2025.acl-long)

Copied to clipboard

Challenge: Large language models have demonstrated remarkable performance across a wide range of language tasks due to their remarkable ability in context modeling.
Approach: They propose to use parallel context encoding to reduce attention entropy by incorporating attention sinks and selective mechanisms to reduce irregular attention . they also propose to incorporate attention sink mechanisms into the parallel encoded context to reduce the irregular attention.
Outcome: The proposed methods lower irregular attention entropy and narrow performance gaps.
Reasoning with OmniThought: A Large CoT Dataset with Verbosity and Cognitive Difficulty Annotations (2026.acl-long)

Copied to clipboard

Challenge: Existing resources often fail to provide extensive reasoning problems with coherent CoT processes distilled from multiple teacher models.
Approach: They propose a large-scale dataset featuring 2 million CoT processes generated by multiple powerful LRMs.
Outcome: The proposed dataset features 2 million CoT processes and is validated by multiple powerful LRMs.
ConvLab-2: An Open-Source Toolkit for Building, Evaluating, and Diagnosing Dialogue Systems (2020.acl-demos)

Copied to clipboard

Challenge: ConvLab-2 inherits Convlab's framework but integrates more powerful dialogue models and supports more datasets.
Approach: They present ConvLab-2, an open-source toolkit that enables researchers to build task-oriented dialogue systems with state-of-the-art models and perform an end-to-end evaluation.
Outcome: The new tool inherits ConvLab's framework and extends it by integrating many recently proposed state-of-the-art dialogue models.
CMOT: Cross-modal Mixup via Optimal Transport for Speech Translation (2023.acl-long)

Copied to clipboard

Challenge: Existing methods to translate speech signals into text are limited by the modality gap between speech and text.
Approach: They propose Cross-modal Mixup via Optimal Transport to overcome the modality gap between speech and text by finding alignment between modalities.
Outcome: Experiments on the MuST-C ST benchmark show that CMOT achieves an average BLEU of 30.0 in 8 translation directions, outperforming previous methods.
BioT5+: Towards Generalized Biological Understanding with IUPAC Integration and Multi-task Tuning (2024.findings-acl)

Copied to clipboard

Challenge: BioT5+ is an extension of the BioT5, but lacked a nuanced understanding of molecular structures.
Approach: They propose a new bio-entity modeling framework, BioT5+, which integrates IUPAC names and molecule data.
Outcome: The proposed model bridges the gap between molecular representations and textual descriptions and improves the grounded reasoning of bio-text and bio-sequences.

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