About Me

I am a PhD student in the Department of Computing at Imperial College London, co-supervised by Prof. Abbas Edalat and Prof. Jiankang Deng.

My research is driven by the intersection of Computer Vision and 3D Graphics, specifically focused on 2D & 3D Face Generation, 3D face reconstruction, talking head generation, and digital humans. I work on generative models (including diffusion models and 3D Morphable Models) for realistic face synthesis, reenactment, and fine-grained facial dynamics control.

Prior to my PhD, I received my MEng in Materials Science and Engineering from Imperial College London (2019–2023), transitioning from biomedical data science (graph neural networks, knowledge graphs) and protein design with geometric deep learning at the Data Science Institute.

Computer Vision 3D Reconstruction Talking Head Generation Diffusion Models 3D Morphable Models Large Language Models

News

Oct 2026
πŸ“„ I'm giving an oral presentation on "Embodied Empathy" at ICMI 2026 LaugHSMI workshop in Napoli, Italy.
Sept 2026
🎀 I'm giving an oral presentation on "MoCoTalk" at IJCB 2026 in Rome, Italy.
July 2026
πŸŽ‰ "MoCoTalk" was accepted to the IJCB 2026!
July 2026
πŸŽ‰ "Embodied Empathy" was accepted to the ICMI 2026 LaugHSMI workshop!

Publications

MoCoTalk: Multi-Conditional Diffusion with Adaptive Router for Controllable Talking Head Generation

Xinyan Ye, Jiankang Deng, Abbas Edalat

IJCB 2026 IEEE International Joint Conference on Biometrics (IJCB 2026)  Β·  Rome, Italy

A multi-conditional video diffusion framework that unifies four complementary control signals β€” reference image, facial keypoints, 3DMM-rendered shading meshes, and speech audio β€” with an Adaptive Multi-Condition Router for fine-grained, controllable talking-head synthesis.

Embodied Empathy: A Multimodal AR and LLM-Powered System for Self-Attachment Psychotherapy with Self-Initiated Humour

Xinyan Ye, Gwyneth Phang, Anandha Gopalan, Abbas Edalat

ICMI 2026 Companion Publication of the 28th International Conference on Multimodal Interaction  Β·  Napoli, Italy

A novel multimodal application operationalising the Self-Initiated Humour Protocol within the Self-Attachment Technique framework. It integrates customisable 3D childhood avatars, augmented reality, and an LLM-driven virtual therapist for automated emotion mirroring in digital mental health tools.

Uplift Modeling based on Graph Neural Network Combined with Causal Knowledge

Haowen Wang, Xinyan Ye, Yikang Wang, Yangze Zhou, Zhiyi Zhang, Longhan Zhang, Jing Jiang, Yiteng Zhai

IEEE CAI 2024 2024 IEEE Conference on Artificial Intelligence, pp. 1487-1492

A scalable uplift modeling framework combining Graph Convolutional Networks (GCNs) with causal knowledge via Conditional Average Treatment Effect (CATE) and adjacency structure learning to effectively estimate treatment effects.

Multihead Causal Distilling Weighting Is All You Need for Uplift Modeling

Haowen Wang, Xinyan Ye, Zhiyi Zhang, Yikang Wang

HPBD&IS 2022 2022 International Conference on High Performance Big Data and Intelligent Systems (HDIS)

Introduces Multihead Causal Distilling Weighting (MCDW) to capture heterogeneous treatment effects and isolate confounding variables from observational data for uplift modeling.

Education & Experience

PhD Researcher in Computing

Imperial College London Β· Department of Computing

2023 – Present

Research on 2D & 3D face generation, 3D face reconstruction, and controllable talking-head synthesis. Co-supervised by Prof. Abbas Edalat & Prof. Jiankang Deng.

MEng Materials Science and Engineering

Imperial College London

2019 – 2023

Master thesis on geometric deep learning for protein design using rotation-invariant spherical harmonics. Supervised by Prof. Stefano Angioletti-Uberti.

Winner Β· Biomedical Data Science Challenge

Imperial College Data Science Institute

2021

Developed an API and web application for brain tumor MRI segmentation.

Lightning Talk Β· Learning on Graphs Conference

Cambridge Meetup Β· Organised by Prof. Petar VeličkoviΔ‡ (DeepMind)

Dec 2022

Presented work on fusing causality and graph neural networks for improved interpretability in representation learning.

Top 10 Β· AI for Sustainable Development Goals Competition

Tsinghua University & United Nations Development Programme (UNDP)

2022

Led a team designing a chatbot for mental illness detection, achieving top 10 out of 200+ teams.