Scalable continuous Machine Learning for non-Stationary Systems

Nonlinear time series prediction attracts researchers from many different disciplines such as engineering systems condition monitoring and control, financial markets prediction and energy management, etc.

Machine learning (ML) is the study of computer algorithms that can improve automatically through experience and based on sample training data, often using a neural network of other modeling paradigms.

One long standing problem of machine learning, known as catastrophic forgetting, is the tendency of the neural network to forget previously learned information abruptly upon learning new information. This makes modeling of nonstationary systems be particularly challenging. There are recent works, called continuous learning or life-long learning, which tackle the problems from diverse aspects for understanding and evaluation, but much is still needed to be done.

The aims of this PhD project are to explore (i) new model topologies; (ii) new learning algorithms and (iii) new industrial applications of continuous learning. The proposed methods should retain modeling capabilities of old tasks while sequentially learn from new tasks and are scalable without need of storing and access increasingly large data sets.

Eligibility requirement: a good Bachelor's degree (2.1 or higher, or equivalent) or Master's degree in computer Science, engineering, mathematics or a similar relevant subject. Industrial experience involving deep learning, data science and its applications is beneficial.

Further enquiries contact: Professor Xia Hong ([Email Address Removed])