ICT4SS - ICT FOR SMART SOCIETIES
Deep neural networks for adversarial generative models on graphs
keywords DEEP LEARNING, VIDEO ANALYSIS
Reference persons ENRICO MAGLI
Thesis type RESEARCH
Description Convolutional neural networks (CNNs) have enjoyed great success in tasks such as image classification, object detection, etc thanks to weight sharing, invariances to geometric transformations and hierarchical
decompositions induced by the convolution operations. Recently, generative adversarial networks (GANs) have used CNNs to learn to generate new images by learning a mapping between a latent space where
random vectors are generated and a "real" image. Some techniques like Bidirectional GANs have been studied to learn the inverse mapping as well, i.e. from an image to the latent space because such latent space has powerful semantic representations of images. At the same time, many data of interest are not defined on regular grids like images, but on generic graphs. It is possible to extend the definition of convolution to data defined over graphs using techniques in the field of graph signal processing.
In this thesis the candidate will extend the bidirectional GANs to data defined on graphs using the graph convolution, implementing the model in Tensorflow. Potential applications are predicting links in social
network graphs or protein interactions in genomics.
Required skills Candidate students should have a background in ICT or mathematics. Knowledge of neural networks (including TensorFlow environment and Python programming) is not a prerequisite, although it would help in the initial stage of the activity.
Deadline 20/11/2018 PROPONI LA TUA CANDIDATURA