These are my cats.
ZHAO Tianyu 趙天雨
Address: Shibuya 2-24-12 39F WeWork, Shibuya-ku, Tokyo 150-6139, JAPAN
E-mail: zhaoty.ting[at]gmail.com
Updated on 2024/03/26
I am a researcher at rinna Co., Ltd.
My research topics include Alignment, Dialogue System, Large Language Model.
General NLP+ML is a part of my interest as well.
I have led the development of rinna's Japanese language models
the qwen-based nekomata series (
HF model page
),
the llama2-based youri series (
HF model page
),
the en-ja bilingual 4b gpt-neox series (
HF model page
),
the ja 3.6b gpt-neox series (
HF model page
),
and earlier models (
HF model page
).
Work Experience
Researcher @ rinna Co., Ltd.
2020.10 - Present
| Alignment, Dialogue, and LLM
Research SDE Intern @ Microsoft Development, rinna Team
2019.10 - 2019.12
| Pre-trained models for Japanese dialogues
RA @ Kyoto University, ERATO Symbiotic HRI Project
2015.12 - 2020.9
| Dialogue system
Education Experience
Ph.D. @ Kyoto University
2017.10 - 2020.9
| Intelligence Science and Technology
| Supervisor: Tatsuya Kawahara
M.Eng. @ Kyoto University
2015.10 - 2017.9
| Intelligence Science and Technology
| Supervisor: Tatsuya Kawahara
B.Sc. @ Peking University
2011.9 - 2015.7
| Computer Science and Technology
| Supervisor: Yunfang Wu
Publications
Enhancing personality recognition in dialogue by data augmentation and heterogeneous conversational graph networks
| Yahui Fu, Haiyue Song, TyZ, and Tatsuya Kawahara
| arxiv Long paper at IWSDS 2024
--- 2023 ---
An integration of pre-trained speech and language models for end-to-end speech recognition
| Yukiya Hono, Koh Mitsuda, TyZ, Kentaro Mitsui, Toshiaki Wakatsuki, and Kei Sawada
| arxiv model code (2023)
Focused prefix tuning for controllable text generation
| Congda Ma, TyZ, Makoto Shing, Kei Sawada, and Manabu Okumura
| arxiv Short paper at ACL 2023
--- 2022 ---
End-to-end text-to-speech based on latent representation of speaking styles using spontaneous dialogue
| Kentaro Mitsui, TyZ, Kei Sawada, Yukiya Hono, Yoshihiko Nankaku, and Keiichi Tokuda
| arxiv demo Regular paper at Interspeech 2022
--- 2021 ---
Multi-referenced training for dialogue response generation
| TyZ and Tatsuya Kawahara
| PDF code Long paper at SIGDIAL 2021
--- 2020 ---
Utterance abstraction and response diversity for open-domain dialogue systems
| TyZ
| PDF Ph.D. thesis
Designing precise and robust dialogue response evaluators
| TyZ, Divesh Lala, and Tatsuya Kawahara
| PDF code Short paper at ACL 2020
Topic-relevant response generation using optimal transport for an open-domain dialog system
| Shuying Zhang, TyZ, and Tatsuya Kawahara
| PDF Regular paper at COLING 2020
--- 2019 ---
Improved end-to-end speech emotion recognition using self attention mechanism and multitask learning
| Yuanchao Li, TyZ, and Tatsuya Kawahara
| PDF non-official code Regular paper (oral) at Interspeech 2019
Effective incorporation of speaker information in utterance encoding in dialog
| TyZ and Tatsuya Kawahara
| arxiv code (2019)
Content word-based sentence decoding and evaluating for open-domain neural response generation
| TyZ, Shinsuke Mori, and Tatsuya Kawahara
| arxiv (2019)
Joint dialog act segmentation and recognition in human conversations using attention to dialog context
| TyZ and Tatsuya Kawahara
| Link code Journal paper at Elsevier CSL 2019
--- 2018 ---
A unified neural architecture for joint dialog act segmentation and recognition in spoken dialog system
| TyZ and Tatsuya Kawahara
| PDF Long paper at SIGDIAL 2018
--- 2017 ---
Joint learning of dialog act segmentation and recognition in spoken dialog using neural networks
| TyZ and Tatsuya Kawahara
| PDF Long paper (oral) at IJCNLP 2017
A conversational dialogue manager for the humanoid robot ERICA
| Pierrick Milhorat, Divesh Lala, Koji Inoue, TyZ, Masanari Ishida, Katsuya Takanashi, Shizuka Nakamura, and Tatsuya Kawahara
| PDF Long paper at IWSDS 2017
--- 2016 ---
Multimodal interaction with the autonomous android ERICA
| Divesh Lala, Pierrick Milhorat, Koji Inoue, TyZ, and Tatsuya Kawahara
| PDF Demo paper at ICMI 2016
Talking with ERICA, an autonomous android
| Koji Inoue, Pierrick Milhorat, Divesh Lala, TyZ, and Tatsuya Kawahara
| PDF Demo paper at SIGDIAL 2016
Interactional and pragmatics-related prosodic patterns in Mandarin dialog
| Nigel Ward, Yuanchao Li, TyZ, and Tatsuya Kawahara
| PDF Regular paper at Speech Prosody 2016
Talks
Nekomata: State-of-the-Art Japanese LLM based on Qwen
2024/01/30
| AI Forward: Alibaba Cloud AI & Big Data Summit 2024 @ Singapore
| Link
日本語LLMの最先端
2023/06/29
| Weights and Biases Tokyo Meetup #5 @ Tokyo
| Link
小冰如何利用FasterTransformer 实现大规模语言模型的产品级部署
2022/12/09
| CNCC 2022 @ Online
| Link
Conversational AI that Learns from the Unverbalized
2021/05/20
| Microsoft Azure AI Days 2021 @ Online