10th February 2021*
Applicants should contact the primary supervisor, and submit their Expression of Interest (EOI) and Application as soon as possible.
*unless filled earlier
Machine learning approaches have been used for developing predictive models such as recommender systems, which seek to predict the preference that a user would give to an item. In recent years a new algorithm named Extreme Learning Machine (ELM) has been developed for training Artificial Neural Networks (ANNs). With ELM, there are no iterations for adjusting connection weights and parameters tuning as in back propagation based ANNs.
While ELM has demonstrated superior performance in developing smaller recommender systems, one drawback of it is that, given an application with a big dataset, the number of neurons in its single hidden layer are typically very large and hence training the network can be computationally impractical. The ELM algorithm’s complexity is at least O(KM2), where K is the number of training instances and M is the number of hidden units. ELM also makes use of batch training, which leads to large memory consumption.
The project aims to evaluate several different solutions (such as representation learning and Deep ELMs) for these problems, and propose a new algorithm for maintaining the strengths of ELM but overcoming its weaknesses in performance and efficiency. Such a solution would be very valuable for developing more effective recommender systems in the current big data era.
The following eligibility criteria apply to this project:
Candidates from a variety of disciplinary backgrounds are encouraged to apply. Knowledge and skills that will be ranked highly include:
Please contact Dr Shuxiang Xu for more information.