Seminar "Research in videogames: use of deep learning for saliency estimation and cheating prevention”

Working in a company that is leader in both the videogames and machine learning fields offers exciting research opportunities at the intersection between these two domains. In this talk Dr. Iuri Frosio (NVIDIA, USA) will illustrate two of these applications. In the first part, the focus will be on saliency prediction in videogames. Based on a dataset of Fortnite sequences acquired in 2019 the peculiar aspects of saliency in videogames will be illustrated, showing that different frames in video sequences are not equally reliable, depending on the number of observers and frame content. A general paradigm that explicitly takes this aspect into account when training a deep learning saliency prediction model is being proposed. In the second part of the talk, the problem of cheating in videogames will be introduced and a deep learning based solution for the case of visual cheating will be proposed. The method allows identifying cheaters with high probability, and it is designed to be robust with respect to potential adversarial attacks put in place by cheat designers. The talk will be concluded by illustrating other potential research areas including videogames and deep learning.