I’m Xiaoyu Shen (沈晓宇), a machine learning scientist at Amazon Alexa AI leading the contextual question-answering project for shopping. I developed the full training pipeline of the system from scratch including pre-training, data collection/annotation and active/continuous learning. The developed shopping QA system has been applied over Amazon apps, webpages and Alexa devices serving hundreds of millions of users.

I obtained my PhD degree from Saarland University/Max Planck Institute for Informatics in Germany under the supervision of Dietrich Klakow and Gerhard Weikum. My research focused on conversational AI, controllable text generation and deep latent variable models. The PhD thesis is available here. During my PhD phase, I also spent 1 year in Japan, working with Prof. Akiki Aizawa from the university of Tokyo/NII, Prof. Kentaro Inui and Prof. Jun Suzuki from Tohuku University/RIKEN AIP. Before that, i obtained my bachelor degree from Nanjing University.

News

Our paper on weak supervision has won the special theme paper award at ACL 2023.
Our paper on cross-lingual product question answering has been accepted by the industry track of ACL 2023. The dataset xPQA has been released here.
We published a survey on training neural-ranking models with limited annotations. Any comments are welcome.

Research Interests

The recent rapid development of LLMs have shown that the model performance can be steadily improved as the growing model and data size. Moving forward, my mission to explore to which extend LLMs can lead us towards the goal and how my expertise could help reduce the gap. With this in mind, I plan on pushing forward in the following research areas:

Interpretability and Explainability: From probabilistic answer generator to logical thinker
Cross-lingual generalization: From English-centric to multilingual expert
Domain Specialization: From general to domain expert

I believe that future LLMs show follow these three paths to become truly usable and reliable intelligent agents. The detailed research statement can be found here.

Awards

Special Theme Paper Award at ACL 2023
Chinese government award for outstanding students abroad, 2020
Best Demo paper Award at COLING 2020
Google NLP Summit Grant, 2019
International Max Planck Research School Fellowship, 2017
Interspeech Travel Grant, 2017
Outstanding Graduate/Outstanding Bachelor Thesis of Nanjing University, 2015
National Scholarship and Outstanding Student of Nanjing University, 2013

Mission

To ordain conscience for Heaven and Earth 为天地立心
To secure life and fortune for the people 为生民立命
To carry on lost teachings of ancient sages 为往圣继绝学
To build peace for posterity 为万世开太平

The development of AI can benefit a wide range of areas such as archeology, education, linguistics, physics, pshychology, etc. Our mission would be to properly guide the development of AI and better contribute to the society. To do so, we need to deeply understand the social needs. I tried to participate in various such as distrubuting food to homeless for the rise foundation, tutoring refugees at the Redi-School, holding anti-separation activities at Bosnia and Herzegovina, etc, and saw what we take for granted every day might be a luxury for many others. Therefore, I strongly believe that AI should be a tool for social good rather than for personal gain. Feel free to contact me for interdisciplinary research that can bring positive social impact and promote general equality.