CV
Education, publications, research, and industry experience.
General Information
| Full Name | Zhiyu Wang (王智语) |
| Research Focus | AI for Biology, Large Language Models, AI Agents, Geometric Deep Learning |
| zhiyu_wang_work@outlook.com |
Education
-
Oct 2024 – Jul 2025 Cambridge, UK
M.Phil. in Advanced Computer Science (Distinction, Rank 8/60)
University of Cambridge - Courses: Geometric Deep Learning, Affective AI, Mobile & Wearable ML, NLP, Multi-agent RL.
-
Sep 2021 – Jun 2024 London, UK
B.Sc. in Computer Science (First Class Honors)
University College London - Courses: Machine Learning, Computer Vision, Reinforcement Learning, Mathematics & Statistics, Software Engineering.
Publications
-
2026 Fast and Interpretable Protein Substructure Alignment via Optimal Transport
ICLR 2026 - Zhiyu Wang*, Bingxin Zhou*, Jing Wang, Yang Tan, Weishu Zhao, Pietro Liò, Liang Hong. * Equal contribution.
-
2025 Topotein: Topological Deep Learning for Protein Representation Learning
arXiv preprint - Zhiyu Wang, Arian Jamasb, Mustafa Hajij, Alex Morehead, Luke Braithwaite, Pietro Liò.
-
2025 MoRE-GNN: Multi-omics Data Integration with a Heterogeneous Graph Autoencoder
arXiv preprint - Zhiyu Wang, Sonia Koszut, Pietro Liò, Francesco Ceccarelli.
-
2025 GraphAU-Pain: Graph-based Action Unit Representation for Pain Intensity Estimation
IJCAI 2025 MiGA Workshop - Zhiyu Wang, Yang Liu, Hatice Gunes.
Research Experience
-
Apr 2025 – Sep 2025 Cambridge, UK
Protein Substructure Alignment via Optimal Transport
University of Cambridge & SJTU - External collaboration | Supervisors: Prof. Pietro Liò & Prof. Liang Hong
- Developed PLASMA, an optimal-transport module that adds fast, highly accurate local-substructure identification to protein representation models with minimal or no training. Accepted as a poster at ICLR 2026.
- Evaluated seven protein-language-model backbones across the VenusX motif, active-site, and binding-site tasks, improving ROC-AUC by 10–30% across all backbones and tasks.
- Used PyMOL case studies to show that retrieved substructures were structurally similar rather than artifacts of sequence similarity.
-
Oct 2024 – Sep 2025 Cambridge, UK
Hierarchical Protein Representation Learning
University of Cambridge - Research Assistant | Supervisor: Prof. Pietro Liò
- Developed TCPNet, an SE(3)-equivariant topological neural network that jointly learns residue- and secondary-structure-level protein features without losing geometric information.
- Received the Highly Commended M.Phil Project Prize 2024–2025 from the Department of Computer Science and Technology.
- Improved performance by 5% on average across three protein-representation tasks when only structural features were provided.
-
Feb 2025 – Sep 2025 Cambridge, UK
Multi-Omics Integration with GNNs
University of Cambridge - Research Assistant | Supervisor: Prof. Pietro Liò
- Used a heterogeneous graph network to embed single-cell RNA, protein, and ATAC modalities into a shared latent space for cell clustering.
- Trained on BM-CITE and LUNG-CITE in under an hour on a laptop GPU, with ARI 3% and NMI 2% below the statistical method MOJITOO.
-
Oct 2024 – Jan 2025 Cambridge, UK
Graph-based Pain Estimation from Facial Expressions
University of Cambridge - Research Assistant | Supervisor: Prof. Hatice Gunes
- Presented at the IJCAI 2025 MiGA Workshop.
- Proposed GraphAU-Pain, combining a ResNet-50 backbone with graph-based facial-action-unit representations for interpretable pain-intensity estimation.
- Developed a transfer-learning strategy for domain differences and missing annotations, improving AU-occurrence F1 by 30% on UNBC.
-
May 2023 – Jun 2024 London, UK
Readmission Prediction with Remote Patient Monitoring Data
University College London - Research Assistant | Supervisor: Prof. Ivana Drobnjak
- Collaborated with UCL Hospital clinicians on a SQL data pipeline, clinical feature engineering, and interpretable ML models (ROC-AUC 0.79 ± 0.03) with SHAP analysis.
- Designed patient-independent stratified bootstrapping for 55 patients with class imbalance, preventing leakage and improving the reliability of deployment estimates.
Industrial Experience
-
Sep 2025 – Present Beijing, China
AI Scientist & Timeseries Generation/Prediction
Sapient Intelligence - ML Research Scientist
- Developing Praxit, an autonomous AI-scientist framework with DAG-based execution lineage for solution provenance; achieved roughly 30% higher performance at 10% of the cost of Claude Code with Opus on Karpathy's AutoResearch prompt.
- Showcased Praxit at AI4 2026 through live booth demonstrations and conversations with prospective customers and collaborators.
- Generating synthetic time-series data and developing HRM-based forecasting models for weight-management applications.
-
Jul 2024 – Sep 2024 Hefei, China
Healthcare LLM Knowledge Distillation
Xunfei Healthcare Technology - Intern ML Researcher
- Built a teacher-LLM knowledge-distillation pipeline that synthesized medical instructions and improved Xiaoyi's instruction-following accuracy from 34% to 82% through supervised fine-tuning.
-
May 2023 – Aug 2023 Shanghai, China
RAG-Based Financial Intelligence System
Luojin Data Information - Intern ML Engineer
- Developed a RAG pipeline with a MongoDB vector store and LangChain agents for automated financial-report classification, reaching 92% query-translation accuracy and a 97% entity-matching rate.
-
Sep 2022 – May 2023 London, UK
Low-cost AI Chatbot Generation System for Hospitals
National Health Service - Intern Software Engineer
- Created an automated hospital-chatbot generation service that reduced development time from weeks to 30 minutes.
- Demonstrated the system to clinicians at Great Ormond Street Hospital and presented it at UCL to IBM, Microsoft, and Intel.
Additional Experience
- Reviewer: NeurIPS 2026; AI4Science Workshop at NeurIPS 2025 and ICML 2026.
- Teaching: Programming Tutor at University College London (2022–23).