Congratulations to Xiangjian Jiang, who has been awarded the Google PhD Fellowship. He is the first recipient from the Department of Computer Science and Technology to receive this honour. The Google PhD Fellowship recognises exceptional research in computer science and provides funding and academic support over a two-year period.
Xiangjian is currently a PhD student in Computer Science at the Cambridge Computer Laboratory, where he is supervised by Professor Mateja Jamnik. His research sits at the intersection of structured tabular data and explainable artificial intelligence (XAI), with a focus on improving the reliability, transparency, and practical deployment of AI systems working with tabular data.
He reflects on his time at Cambridge and how he hopes his research will contribute to the future of AI.
Your journey to Cambridge
Before Cambridge, I became increasingly drawn to real-world data problems where reliability matters. What ultimately brought me here was recognising that much of the world’s most important information – such as in healthcare and finance – lives in tables, yet the AI field has historically paid less attention to them.
What drew you to pursue advanced study in computer science, and how did you develop a fascination for foundational AI research?
I wanted to work on questions that sit beneath individual model tweaks: how we build models that genuinely understand data structures and generalise robustly. I developed a fascination with foundational AI research through the gap between mainstream breakthroughs (text and images) and the structured, heterogeneous data that drives real decisions, i.e., tabular data.
Why did you choose to study at St Edmund’s?
I chose to study at St Edmund’s because I completed my MPhil here and really valued the atmosphere of a mature college. It’s a place where I can meet people from very different backgrounds and disciplines, which I find both grounding and intellectually refreshing alongside my research.
You’ve spoken about feeling well supported by the College during your MPhil and PhD. What aspects of the College’s community have shaped your academic experience the most?
The diversity of backgrounds has also been unusually valuable: conversations at Eddies are often with people who think in completely different frameworks – medicine, policy, humanities, engineering – and that has helped me pressure-test my ideas and communicate my work more clearly. I have appreciated the practical side: the College feels well set up for graduate life, and that reliability reduces friction during intense research periods.
What first sparked your interest in tabular data research, an area you describe as “unloved” within the AI community?
My interest started from healthcare. My father is a doctor specialising in diabetes, and diabetes diagnosis relies on many tests that are recorded as tables. That made it very clear to me that tabular data is central to high-impact decision-making. At the same time, it felt “unloved” compared with text and vision, so I wanted to help build the kind of large, capable tabular models that could one day be as important as large language models.
How did you feel when you learned you had been awarded the Google PhD Fellowship?
I felt it as a major confidence boost, real moral support. It was reassuring to see external recognition that this research direction matters and isn’t only something I personally find important.
This Fellowship recognises pioneering work and comes with significant academic support. How do you see it influencing your research direction over the next two years?
It gives me practical support: funding, computational resources, and connections with other researchers, which I expect will accelerate my work on tabular foundation models. It also helps build momentum around the area: instead of me always having to persuade people that tabular modelling is important, more researchers may become interested and reach out.
You’ve highlighted potential impact in healthcare and finance. Can you share an example of how your work could make a meaningful difference in the real world?
For instance, the first piece of work during my PhD journey, ProtoGate [1], which is an instance-wise based feature selection method for any tabular data. In healthcare, people usually want tabular models that can analyse historical disease and treatment records and help predict which therapies are likely to lead to good patient outcomes. In finance, tabular models could better interpret structured signals—like company fundamentals or market indicators—while remaining robust and reliable when the data distribution shifts.
[1] Jiang, Xiangjian, et al. “ProtoGate: prototype-based neural networks with global-to-local feature selection for tabular biomedical data.” Proceedings of the 41st International Conference on Machine Learning. 2024.
What has been the most challenging aspect of developing a new approach to AI modelling, and what continues to motivate you?
The most challenging part is that tabular data is expressed very differently from language: it’s heterogeneous, context-dependent, and structured by columns that vary widely across datasets. We can’t just adapt a language model to “read a spreadsheet” and expect dependable reasoning. That motivates my focus on learning the underlying relationships, especially causal structure, using statistical techniques, so the models are less likely to be misleading for practitioners.
Can you share a little bit about your first impressions of both Cambridge and St Edmunds?
Cambridge initially felt both inspiring and demanding: the concentration of expertise is obvious, and the academic pace can be relentless. At the same time, I found the culture to be unusually serious about ideas: people engage with questions rigorously, and that quickly raises your own standards. My first impression of St Edmund’s was different in a good way: it felt calm, welcoming, and internationally minded, with a mature atmosphere where it is normal to be at different stages of life and career. That combination – high intensity in the University, steadier rhythm in the College – ended up being a good fit for how I work.
How do you hope your research will contribute to the future of AI, especially in fields currently underserved by mainstream model development?
I want to help push foundational AI beyond text and images by developing models that can genuinely understand, reason over, and make reliable predictions from tabular data. Because tabular data dominates many high-stakes domains, I also want to raise the standard for trustworthiness by focusing on underlying relationships—including causal relationships—rather than surface-level pattern matching.
What are your aspirations after completing your PhD?
After my PhD, I want to continue doing research that sits at the boundary of foundations and application—building reliable machine learning systems for structured/tabular data, and setting standards for how we evaluate and trust them.
What advice would you give prospective international students thinking of choosing St Edmund’s College?
If you are an international student, I would prioritise two things: whether you will have a stable, supportive base outside your department, and whether the community will broaden your perspective. St Edmund’s is particularly strong on both because it is mature student-focused (St Edmund’s accepts students aged 21 and above) and genuinely diverse. My practical advice is to be proactive early: introduce yourself, attend a few events even when you are busy, and build a small support network in College as well as in your lab. That community becomes especially valuable during high-pressure periods of research and writing.
Are you an international student considering studying at Cambridge? Find out more about Life in the UK.
