Z-ENG: Transformer based Recommender
The primary goal of recommender systems is to estimate user preferences on items, e.g., audio, video content. The transformer based technique is intended to be the replacement of traditional RNN architectures, as LSTM and GRU. The LSTM/GRU based recommendation can be treated as a prediction task, e.g., personalized playlist generation. The student will work on the elaboration and evaluation of such a technique. For details on predictive techniques, see: https://arxiv.org/pdf/2004.13715.pdf, Section 4.1.