Kaitao Song

About

Kaitao Song is a senior researcher at Microsoft Research Asia, Shanghai, China. His research interests focus on machine learning/deep learning algorithms on natural language processing/speech processing, including pre-trained language model, neural machine translation, music generation, text summarization, neural architecture search for NLP, audio speech recognition,  text-to-speech synthesis and etc.

He has published multiple famous papers, including:

• HuggingGPT / JARVIS
  • Github (opens in new tab) stars: 23.1K
  • The Most Influential Papers in NeurIPS 2023, Rank 8
  • integrated into Langchain (opens in new tab) and HuggingFace (opens in new tab)
  • NeurIPS 2023 Highlights (opens in new tab)
  • Open100: Top 100 Open Source achievements (opens in new tab)
• PVT
  • The Most Influential Papers in ICCV 2021, Rank 2
• MPNet
  • NeurIPS 2020, integrated into HuggingFace (opens in new tab) and SBert (opens in new tab)
  • Leading first in sentence similarity (opens in new tab)
  • Top-10 downloads in Hugging Face models (opens in new tab)
• MASS
  • The Most Influential Papers in ICML 2019, Rank 9

Currently, He is focused on the following research topics:

• Large Language Models
  • Autonomous Agents, Planning, Tool Use, Memory and etc
• General NLP methods
  • Pre-trained Model, Architecture Design, Efficient Training and Inference.

Hiring: We are looking for research interns with self-motivation. Please contact me (kaitaosong[AT]microsoft.com) if you are interested in my research topics.

Research Topics & Projects

• Large Language Models
  • Autonomous Agents: HuggingGPT (opens in new tab), MusicAgent (opens in new tab)
    • Tool: EasyTool (opens in new tab)
    • Benchmark: TaskBench (opens in new tab)
  • Prompt Engineering: DTG (opens in new tab), EvoPrompt (opens in new tab)
• Foundation Models
  • Language: MASS (opens in new tab), MPNet (opens in new tab), Transcormer (opens in new tab)
  • Vision: PVT (opens in new tab), PVT v2 (opens in new tab)
  • Compression: LightPAFF (opens in new tab), NAS-BERT (opens in new tab), MAGIC-NAS (opens in new tab)
• NLP Application
  • Translation: [Paper-1 (opens in new tab)], [Paper-2 (opens in new tab)], [Paper-3 (opens in new tab)]
  • Summarization: [Paper-1 (opens in new tab)]
  • Information Extraction: DiffusionNER (opens in new tab)
  • Music: SongMASS (opens in new tab), DeepRapper (opens in new tab)
  • MultiModal: [Paper-1 (opens in new tab)], [Paper-2 (opens in new tab)]
• Speech Application
  • Clinical Application: [Paper-1 (opens in new tab)], [Paper-2 (opens in new tab)], [Paper-3 (opens in new tab)], [Paper-4 (opens in new tab)]
  • ASR & TTS: [Paper-1 (opens in new tab)], [Paper-2 (opens in new tab)], [Paper-3 (opens in new tab)]

Recent News:

[2024-05-15] Our paper (NaturalSpeech 3) has been accepted by ICML 2024. Our papers about LLM reasoning has been accepted by ACL 2024.

[2024-02-29] I have given a talk about HuggingGPT (opens in new tab) at AGI Leap Summit 2024.

[2024-01-17] Two our papers (EvoPrompt, VoiceGen) have been accepted by ICLR 2024. And Two of our papers (EasyTool, TaskBench) also have been accepted by ICLR 2024 LLMAgent workshops.

[2023-12-10] I have given a talk – “The Future is Here – A Deep Dive into Autonomous Agents (opens in new tab)” at NeurIPS 2023

[2023-09-22] HuggingGPT has been accepted by NeurIPS 2023