Bayesian Deep Learning for Deep Reinforcement Learning – We present an algorithm for learning to move to an unknown location, in the case that it is too challenging to follow rules. We show that the number of possible directions is exponentially large when the number of possible actions is at least two orders of magnitude. We also provide a mechanism for automatically learning to move to the unknown location by estimating a probability distribution over the environment. Our results reveal that the optimal behaviour of a general-purpose deep convolutional neural network may be reduced to the task of estimating the location of a robot by using this distribution alone and further showing how this procedure can improve the quality of navigation by taking into account its own behaviour and its own uncertainty.

We consider the problem of identifying movements from unstructured data, and provide a simple implementation on mobile devices in the domain of robotics. To this end, we provide a real-time interactive platform to users (using an embedded computer) able to perform movement identification in real-time and control robot vehicles in real time while maintaining safety and navigation. Our platform provides users an opportunity to access these skills through the interactive robotic interaction, and is the first such platform for autonomous mobility of robots using real-time interactive control and navigation.

Modeling the results of large-scale qualitative research using Bayesian methods

A Non-Parametric Graphical Model for Sparse Signal Recovery

Bayesian Deep Learning for Deep Reinforcement Learning

Learning Visual Representations by Mining Object and Category Similarities

The Classification, GAN and Supervised Learning of Movement Recognition SystemsWe consider the problem of identifying movements from unstructured data, and provide a simple implementation on mobile devices in the domain of robotics. To this end, we provide a real-time interactive platform to users (using an embedded computer) able to perform movement identification in real-time and control robot vehicles in real time while maintaining safety and navigation. Our platform provides users an opportunity to access these skills through the interactive robotic interaction, and is the first such platform for autonomous mobility of robots using real-time interactive control and navigation.