Bayesian deep learning via MCMC sampling with application to robot path planning

QUT Centre for Robotics
13th September 2022
Bayesian models provide a probabilistic representation of parameters that have an advantage over single-point estimates with uncertainty quantification in model predictions. Markov Chain Monte Carlo (MCMC) sampling methods are typically used to implement Bayesian inference for estimating and uncertainty quantification of parameters in models. The progress of MCMC sampling for neural networks and deep learning models has been slow due to the curse of dimensionality, and multi-modal posterior distributions. Recent advances in parallel computing and proposal schemes in Markov Chain Monte Carlo (MCMC) sampling have opened the path for Bayesian deep learning. This seminar provides an overview of the implementation of Bayesian deep learning methods that begin with simple neural networks and extends to deep learning models such as autoencoders. It provides a roadmap of Bayesian inference via MCMC for uncertainty quantification for robot path planning with Lyapunov-based control that can be extended for drones.
About the Presenter:
Dr Rohitash Chandra is a Senior Lecturer in Data Science at the UNSW School of Mathematics and Statistics. Dr Chandra leads a program of research encircling methodologies and applications of artificial intelligence; particularly in areas of Bayesian deep learning, neuro-evolution, climate extremes, geoscientific models, and mineral exploration. Dr Chandra has developed novel methods for machine learning inspired by neural systems and learning behaviour that include transfer and multi-task learning, with the goal of modular deep learning. His current interests are ensemble learning, data augmentation, applied language models, bioinformatics, and machine learning for COVID-19. research.unsw....
Organised by Prof. Will Browne and Dr Dorian Tsai
Download presentation: github.com/roh...