Identifying Structure in Multitask Learning


Carl Allen

Whilst modern machine learning methods achieve impressive results across a wide variety of tasks, from image and speech recognition to playing games, an algorithm able to efficiently learn a wide variety of tasks – as humans do – remains elusive. Such an algorithm would leverage acquired knowledge to more efficiently learn new tasks without detriment to those already learned. More than simply learning a growing catalogue of tasks, this amounts to abstraction of the learning process itself, or learning to learn.

Supervisors: Tim Hospedales & Iain Murray 


Previous: Intelligent Exploration with Bayesian Conditional Density Estimation            Next : Meta-Learning approaches in Deep Neural Networks