Training Deep Neural Networks via Multi-task Optimisation

Dr Zhang’s thesis work contributes to a novel way to effectively training deep neural networks which are the mainstay of modern AI techniques, aiming to improve the less desirable generalisation performance achieved by using traditional approaches to train deep neural networks due to inherent ambiguity of generalisation formulation. It will benefit a variety of AI systems in many scientific and engineering fields that employ deep neural networks as their backbones.