Shaobo Cui

I am a fourth-year PhD student in Computer Science at EPFL, working under the supervision of Prof. Boi Faltings. I am co-supervised by Prof. Antoine Bosselut and mentored by Prof. Michael C. Gastpar. I am expected to complete my PhD by July, 2025.

Previously, I worked as a senior NLP engineer at DAMO Academy, Alibaba Group, where I led projects in machine comprehension, dialogue systems, and recommendation algorithms for digital humans. Notably, I developed the recommendation algorithm for the digital human project from 0-1, leading to the debut of the Virtual Influencer at Olympic Games, Beijing 2022. I also hold 9 patents.

Before joining Alibaba, I earned my Bachelor degree from Beihang University and my Master degree from Tsinghua University.

Research Highlights: Human-Machine Interaction

I am passionate about enhancing machines’ ability to understand and interact with humans:

• Industrial Dialogue Systems: (i) Enhancing retrieval dialogue systems using session information(QSM paper @ CIKM, CSM @ KDD, patent, patent); (ii) Developing methods for generative dialogue systems, including teacher-student(MTSS) framework(AAAI paper) and dual adversarial learning(DAL) framework (workshop@NAACL paper); (iii) Detecting optimal response timing to in industrial dialogue systems(TASLP paper); (iv) Knowledge extraction via a OneStop approach workshop@WWW; (v) Developing methods and platforms for the automated testing of dialogue systems patents.
• Virtual Influencer: (i) Built the recommendation algorithm of Alibaba Virtual Influencer from 0-1, leading to its debut at Olympic Games; (ii) Applying generative methods to assist virtual influencers in constructing their manuscript CIKM paper.
• Machine Learning: Greedy coordinate descent, regularized optimal transport, and Lipschitz constrainted WGAN(NeurIPS paper, ITCAI paper).

Research Highlights: Causality

My PhD thesis aims to advance machine understanding of causality, with a focus on:

• Commonsense Causality: Feel free to check our survey about commonsense causality.
• Defeasibility in Causality: We defined and quantified defeasibility, an often-overlooked aspect of causality (ACL 2024).

• LLMs’ Consistency on Causality: We empirically study whether LLMs can keep self-consistency over nuances hidden inside causality preprint.
• Causality for Science: Ongoing. Will release it later.

For Prospective Students

I have reserved time slots to supervise master students for thesis and semester projects. Undergraduate and master’s students interested in research projects are encouraged to read this page for more information.