Jiahao Yu (俞嘉豪)

GSK Medicines Research Centre

Gunnels Wood Road

Stevenage, SG1 2NY, UK

I am a third year BSc Data Science student at University of Bristol, taking a Computational Scientist Industrial Placement at GSK.

During my placement at GSK, I am co-supervised by Dr. David Marcus (GSK) and Prof. Christophe Andrieu (Bristol). My research areas involve missing data imputation, augmentation, and uncertainty estimation/mitigation, in the setting of drug discovery data.

At University of Bristol, I work with Dr. Song Liu on a part-time basis. My research areas include missing data imputation by utilizing gradient flows.

Previously, I worked with Dr. Mahmood Sharif on adversarial machine learning at Tel Aviv University.

Feel free to contact me at jiahao.yu at bristol.ac.uk for anything.

P.S. I am actively looking for a PhD position in Statistical Machine Learning starting 2025 fall, globally. I am also looking for a research studentship in 2025 summer. I would be greatly appreciated to hear from potential supervisors.

news

May 02, 2024	I am happy to announce that our paper Minimizing f-Divergences by Interpolating Velocity Fields has been accepted for ICML 2024, in Vienna, Austria!

selected publications

ICML

Minimizing f-Divergences by Interpolating Velocity Fields

Song Liu, Jiahao Yu, Jack Simons , Mingxuan Yi , and Mark Beaumont

In Proceedings of the 41st International Conference on Machine Learning , Jul 2024

Abs arXiv PDF

Many machine learning problems can be formulated as approximating a target distribution using a particle distribution by minimizing a statistical discrepancy. Wasserstein Gradient Flow can be employed to move particles along a path that minimizes the f-divergence between the target and particle distributions. To perform such movements we need to calculate the corresponding velocity fields which include a density ratio function between these two distributions. While previous works estimated the density ratio function first and then differentiated the estimated ratio, this approach may suffer from overfitting, which leads to a less accurate estimate. Inspired by non-parametric curve fitting, we directly estimate these velocity fields using interpolation. We prove that our method is asymptotically consistent under mild conditions. We validate the effectiveness using novel applications on domain adaptation and missing data imputation.
UK-QSAR

Filling the gaps: Data Imputation Methods for Drug Discovery

Jiahao Yu, and David Marcus

In UK-QSAR Spring 2024 Meeting , Apr 2024

Abs Poster

Drug discovery datasets are often shown as sparse, noisy, and heterogeneous. To facilitate drug discovery projects and to ensure the effectiveness of Machine Learning (ML) algorithms and predictive models, it is necessary sometimes to find methods that can fill in the gaps in this data. In this poster, we discuss several classic and state-of-the-art data imputation methods and compare their performance with classic QSAR modelling. We found that data imputation models can usually outperform classic QSAR models, however some are unsuitable for data imputation in drug discovery, and some will require extensive calculation time.