Raphael Shu Profile

Bio

Raphael Shu is a Senior Applied Scientist at Amazon AWS GenAI. He received his PhD from the University of Tokyo in 2020, earning a Dean’s Award. His research primarily focuses on Natural Language Generation, Conversational AI, and AI Agents, with publications in conferences such as ICLR, ACL, EMNLP, and AAAI. His work on the attention mechanism and discrete latent variables received an Outstanding Paper Award at ACL 2017 and the Best Paper Award for JNLP in 2018 and 2019. Raphael worked as the Head of Research at Weblio, Inc. in 2015 and continued to serve as an R&D consultant for the CEO until 2020, launching multiple machine learning products, including “AI Eikaiwa”, an AI chatbot for practicing English conversation. At AWS, he led the Dialog2API project, which enables large language models to interact with the external environment through dialogue. Since 2023, he has led a team aiming to develop the Dialog2API capability for Titan, Amazon’s in-house large language model.

Research Interests

• LLM-based Agents
• Reinforcement Learning for Generative Models
• Federated / Distributed Learning

Cool Stuffs

• Still maintaining
  • Deep Learning Monitor
• No longer maintaining
  • tree2code
    • Learning discrete codes for syntax
  • lanmt
    • Non-autoregressive neural machine translation with latent variables and refinement
  • Visualization of optimization methods (Chrome-only)
  • Visualization of Monte Carlo methods (Chrome-only)

Papers

Google Scholar page

Codes

• tree2code: Learning Discrete Syntactic Codes for Structural Diverse Translation
  • https://github.com/zomux/tree2code
• nmtlab: A simplified framework for neural machine translation
  • https://github.com/zomux/nmtlab
• neuralcompressor
  • Compress word embeddings
  • https://github.com/zomux/neuralcompressor
• deepy
  • Highly extensible deep learning framework based on Theano
  • https://github.com/zomux/deepy
• neuralmt
  • Neural machine translation modelling tookit
  • https://github.com/zomux/neuralmt
• deepy-draw
  • An implementation of DRAW with deepy framework
  • 
  • https://github.com/zomux/deepy-draw/
• JRNNLM
  • RNNLM implementation in JAVA
  • https://github.com/zomux/jrnnlm

Links

• Old Bio page in The University of Tokyo