Cyber Security Webinar by Dr. Yunpeng Li - Uncertainty quantification and propagation in high-dimensional spaces

Join Zoom Meeting

https://tudelft.zoom.us/j/99314357068

Meeting ID: 993 1435 7068
Passcode: 678062

 Abstract: 

Deep learning models, which are cornerstones in the success of modern machine learning applications, often lack representation of uncertainty or produce overconfident predictions that can lead to costly consequences. The optimal transport theory has found it application in quantifying uncertainty in large-scale machine learning problems by measuring the Wasserstein distance between data distributions. Separately, for real-world sequential inference applications, differentiable filters provided a mathematically principled framework to propagate model uncertainty. This talk will hence be divided into two parts. In the first part, I will discuss a new family of slice-based Wasserstein distance metrics, called augmented sliced Wasserstein distances (ASWDs), with a novel incorporation of injective neural networks. ASWDs learns nonlinear projections that can capture the complex structure of the data distributions which improve their projection efficiency. In the second part, I will discuss our recent efforts in constructing more expressive dynamic model and proposal distributions in the differentiable particle filtering framework through normalizing flow. In addition, I will introduce an end-to-end learning objective based upon the maximisation of a pseudo-likelihood function which can improve the estimation of states when large portion of groundtruth information are unknown.

Bio:

Dr Yunpeng Li is Senior Lecturer in Artificial Intelligence in the Department of Computer Science at the University of Surrey in the UK. Before joining Surrey as a Lecturer in AI in 2018, he was a postdoctoral researcher in the Machine Learning Research Group in the Department of Engineering Science at the University of Oxford and was a Junior Research Fellow at Wolfson College at Oxford. He received a PhD in Electrical Engineering at the McGill University in Canada in 2017. His research interests are in the areas of statistical machine learning and signal processing, particularly Bayesian inference techniques and the optimal transport theory. He has broad interests in applications of machine learning, e.g. breast cancer detection, dental disease detection, and environment acoustic sensing. His work has won the best paper award in the NeurlPS Workshop on Machine Learning for Developing World in 2018 and 2019.