Deep and Multi-fidelity learning with Gaussian processes

Time: 14:00 - 15:00

Venue: Room 3.009 Alliance Manchester Business school, Booth Street West, Manchester, M5 6PB

Speaker: Andreas Damianou

Title: Deep and Multi-fidelity learning with Gaussian processes

Abstract:

I will first talk about Deep Gaussian processes (deep GPs), a nested GP model which can learn rich representations in a data efficient manner, while being able to reason about uncertainty. I will then introduce GP multi-fidelity learning as a way of fusing multiple fidelities of the data through Uncertainty Quantification and Emulation. Multi-fidelity methods are prominently used when cheaply-obtained, but possibly biased and noisy observations must be effectively combined with limited or expensive true data in order to construct reliable models. Finally, I will show how a deep GP manifests itself as a natural model for multi-fidelity learning, where uncertainty propagation between layers / fidelity levels is crucial to avoid overfitting.
 

Bio:

I am a Machine Learning scientist at Amazon Research, Cambridge. My research focuses on Bayesian probabilistic modeling as well as on methods for improving deep learning through knowledge transfer and uncertainty quantification. I'm particularly interested in applying my research in decision making systems. In the past I have worked in perceptual models for bio-inspired robotics. I have pursued a PhD degree with Neil Lawrence at Sheffield, developing Deep Gaussian Processes. Website: andreasdamianou.com