Yang Feng is a Professor at the Institute of Computing Technology, Chinese Academy of Sciences (ICT/CAS). After receiving her PhD from ICT/CAS, she worked at the University of Sheffield and the Information Sciences Institute, University of Southern California for the following several years. She currently leads the Natural Language Processing group at ICT/CAS, with research interests spanning machine translation, dialogue and LLMs. Her research now mainly focuses on interactive and streaming multi-modal LLMs, and RAG for LLMs. A recipient of the Best Long Paper Award at ACL 2019, Yang Feng has developed and open-sourced the large language model of BayLing, enhancing the support for non-dominant languages, as well as the interactive speech LLM of LLaMA-Omni. 

• September 26, 2024, our paper of LLM decoding acceleration has been accepted by NeurIPS 2024.

• September 11, 2024, we released LLaMA-Omni, a real time interactive speech LLM which can interact via speech with users at extremely low latency,  like GPT 4o. Paper, code and model are available.

• June 5, 2024, we released StreamSpeech, a simultaneous speech translation model which can perform streaming ASR, simultaneous speech-to-text translation and simultaneous speech-to-speech translation via an "All in One" seamless model. Paper, Code, Models and Demo are available.

• May 2024, We released BayLing · Translate, the first LLM that can support simultaneous machine translation. Welcome to try the LLM of BayLing and BayLing · Translate . Please infer to  the news for more information

• May 2024, We have 9 papers accepted by ACL 2024.

• We have 6 papers accepted by EMNLP 2023.

• We have 4 papers accepted by NeurIPS 2023.

• We released a paper about BayLing, an instruction-following LLM, please refer to  https://nlp.ict.ac.cn/bayling for more details.

• I and Chenze Shao gave a tutorial in EMNLP 2022 about "Non-Autoregressive Models for Fast Sequence Generation".

Selected Publications:  [More Publications]

• Zhuofan Wen, Shangtong Gui, Yang Feng. Speculative Decoding with CTC-based Draft Model for LLM Inference Acceleration. To appear in Proceedings of NeurIPS 2024.

• Qingkai Fang, Shoutao Guo, Yan Zhou, Zhengrui Ma, Shaolei Zhang, Yang Feng. LLaMA-Omni: Seamless Speech Interaction with Large Language Models. In arXiv:2409.06666.

• Shaolei Zhang, Tian Yu, Yang Feng. TruthX: Alleviating Hallucinations by Editing Large Language Models in Truthful Space. In Proceedings of ACL 2024.

• Qingkai Fang, Shaolei Zhang, Zhengrui Ma, Min Zhang, Yang Feng. Can We Achieve High-quality Direct Speech-to-Speech Translation Without Parallel Speech Data?. In Proceedings of ACL 2024.

• Qingkai Fang, Yan Zhou, Yang Feng.  DASpeech: Directed Acyclic Transformer for Fast and High-quality Speech-to-Speech Translation.  NeurIPS 2023.

• Shaolei Zhang, Yang Feng.  Unified Segment-to-Segment Framework for Simultaneous Sequence Generation. NeurIPS 2023.

• Chenze Shao, Zhengrui Ma, Min Zhang, Yang Feng. Beyond MLE: Convex Loss for Text Generation. NeurIPS 2023.

• Shangtong Gui, Chenze Shao, Zhengrui Ma, Xishan Zhang, Yunji Chen, Yang Feng. Non-autoregressive Machine Translation with Probabilistic Context-free Grammar. NeurIPS 2023.

• Shaolei Zhang, Qingkai Fang, Zhuocheng Zhang, Zhengrui Ma, Yan Zhou, Langlin Huang, Mengyu Bu, Shangtong Gui, Yunji Chen, Xilin Chen, Yang Feng.  BayLing: Bridging Cross-lingual Alignment and Instruction Following through Interactive Translation for Large Language Models. In arXiv:2306.10968.

• Yang Feng, Chenze Shao. Non-Autoregressive Models for Fast Sequence Generation. In Proceedings of EMNLP 2022: Tutorial Abstracts.

• Chenze Shao, Yang Feng. Exploring Non-Monotonic Latent Alignments for Non-Autoregressive Machine Translation. NeurIPS 2022 (Spotlight).

• Wen Zhang, Yang Feng, Fandong Meng, Di You, Qun Liu. Bridging the Gap between Training and Inference for Neural Machine Translation. In Proceedings of ACL 2019. (Winnner of ACL2019 Best Paper Award)

 LLaMA-Omni：Real Time Interactive Speech LLM

[Paper] [Code] [Model-8b]

September 11, 2024, we released LLaMA-Omni, an interactive speech LLM which can interact with users at extremely low latency, just like GPT 4o. LLaMA-Omni was trained based on Llama-3.1-8B-Instruct for generic instruction following, working in the manner that accept speech instructions, and respond with text and speech simultaneously.  LLaMA-Omni outperforms the exsiting open-source speech LLMs  in terms of both answer content and answer style, achieving the average  latency of 226 ms, even lower than that of GPT 4o which is 320ms.

StreamSpeech: 实时语音翻译模型

[Paper] [Code] [Models] [Demo]

June 5, 2024, we released StreamSpeech, a simultaneous speech translation model which can perform streaming ASR, simultaneous speech-to-text translation and simultaneous speech-to-speech translation via an "All in One" seamless model. StreamSpeech achieves SOTA performance in both offline and simultaneous speech-to-speech translation on with public test sets.

 BayLing Large Language Model

[Homepage] [Demo] [Paper] [Code] [Model-7b] [Model-13b]

The existing LLMs are usually focused on English, leading to inferior performance in non-English languages. In order to improve the performance for non-English languages, it is necessary to collect language-specific training data for foundation LLMs and construct language-specific instructions for instruction tuning, both of which are heavy loads. To minimize human workload, we propose to transfer the capabilities of language generation and instruction following from English to other languages through an interactive translation task. We have developed BayLing, an instruction-following LLM by utilizing LLaMA as the foundation LLM and automatically constructing interactive translation instructions for instructing tuning. Extensive assessments demonstrate that BayLing achieves comparable performance to GPT-3.5-turbo, despite utilizing a considerably smaller parameter size of only 13 billion.

• On June 21, 2023，we publish the paper of BayLing on Arxiv-- Bridging Cross-lingual Alignment and Instruction Following through Interactive Translation for Large Language Models，open source the code, model-7b and model-13b, and release the online demo as well.

• Principle and Applications of Large Language Models:  Summer 2023, 2024

• Natural Language Processing:  Fall 2022, 2023

• Weichang Qian Chinese Information Processing Science&Technology Award--Hanwang Youth Award First Prize

• "Young Scientist" of Technical Committee on NLP, China Computer Federation 

• "Outstanding Contribution Award" of Chinese Association of Artifical Intelligence

• ICT/CAS “Brilliant Star”, 2019

• Best Long Paper Award of ACL 2019

• ICT/CAS “Baixing Talent Introduction Program”

• Associate Editor: IEEE Transactions on Asian and Low-Resource Language Information Processing（2022-2023);

• Editorial Board Member：Ariticial Intelligence;  The Northern European Journal of Language Technology;

• Senior Area Chair:  EACL 2024;  AACL 2022;  EMNLP2021;

• Area Chair: NAACL 2024; EACL2023;  ACL2021；EMNLP2020; COLING 2018;

• Organizing Committee Member: ACL 2024 Demonstration Co-chair;  ACL 2023 Workshop Co-chair;  EMNLP 2020 Virtual Infra-structure Co-chair;

• Program Co-chair: YSSNLP 2024; CCMT 2023; CCL 2022; Language & Intelligence Summit 2021;

• ACL 2024 Paper Awards Committee member；

• ARR Permanent Senior Action Editor (from June 2022);

• Doctorate Committee Assessor of KU Leuven;

• Dissertation Committee Member of Worcester Polytechnic Institute.

Ph.D Candidates:   

• Shaolei Zhang,   2020-present

• Zhuocheng Zhang,   2020-present (Co-advising with Prof. Min Zhang) 

• Qingkai Fang,   2021-present 

• Zhengrui Ma,   2021-present (Co-advising with Prof. Min Zhang) 

• Shoutao Guo,   2022-present 

• Kehao Zhang,   2022-present

• Mengyu Bu,   2023-present

• Yan Zhou,   2023-present

• Zhuofan Wen, 2024-present

• Jilong Zhu, 2024-present

Master Candidates:    

• Ziwen Yan,   2022-present

• Zhe Yang,   2022-present

• Tian Yu,   2023-present
• Kangyu Qiao, 2024-present

Alumni

Ph.D:   

• Jiao Ou,   Graduated in 2023

• Shuhao Gu,   Graduated in 2023

• Chenze Shao,   Graduated in 2023

• Shuman Liu,   Graduated in 2023 (Co-advising with Prof. Qun Liu from 2022)

• Wen Zhang,   Graduated in 2019 (Co-advising with Prof. Qun Liu from 2017)

Master:   

• Langlin Huang, Graduated in 2024

• Longxiang Liu, Graduated in 2024

• Xuanfu Wu,   Graduated in 2023

• Dengji Guo,   Graduated in 2022

• Zekang Li,   Graduated in 2022

• Jicheng Li,   Graduated in 2022

• Yong Shan,   Graduated in 2021

• Shugen Wang,   Graduated in 2020 (Co-advising with Prof. Qun Liu)

• Jingyu Li,   Graduated in 2019 (Co-advising with Prof. Qun Liu)

• Haiyang Xue,   Graduated in 2019 (Co-advising with Prof. Qun Liu)       .

  
  

  Updated on September 12, 2024.