Jihao Andreas Lin, jal232@cam.ac.uk
Germany
Engineering, Queens' College
PhD thesis title: Uncertainty Quantification in Bayesian Deep Learning
Research interests:
1. Bayesian Machine Learning.
2. Function-space inference.
3. Linear Models and Approximations.
4. Deep Learning and Computer Vision.
Deep learning systems have demonstrated remarkable success at tasks which were historically considered extremely difficult or even impossible for computers, such as computer vision or natural language processing. However, they often exhibit erratic behaviour, particularly in the absence of large amounts of data or when encountering unfamiliar scenarios. Therefore, currently, their use in sensitive environments, like medicine or traffic, is limited. Bayesian deep learning attempts to provide robustness and safety by treating model parameters and predictions as random instead of deterministic variables, such that uncertainties can be quantified to inform decision-making. My research is focused on developing generic algorithms which are capable of producing well-calibrated uncertainty estimates with the goal of opening new avenues for the employment of powerful deep learning systems in sensitive yet highly relevant environments, increasing the effectiveness, efficiency and availability of vital resources and services.
Who or what inspired you to pursue your research interests?
Throughout my academic journey, I developed both a genuine interest in abstract ideas as well as the desire to solve relevant real-world problems. Bayesian deep learning appeals to me because I believe that it can satisfy both at the same time, relying extensively on mathematics while having the potential to revolutionise various critical fields due to the generic nature of learning algorithms.