Implicit graph neural networks for fake news early detection

Description of field of research:

In the past few years, implicit models are emerging structures in deep learning where the outputs are determined implicitly by a solution of some underlying sub-problem. For example, chen et al. [1] model deep ResNets by black-box ordinary differential equations (ODE) solvers in forward and backward passes (as if the network has smaller “layer steps”) given the start- and end-points of a dynamical system. Bai et al. [2] proposed the deep equilibrium approach (DEQ), which models temporal data by directly solving for the sequence-level fixed point and optimising this equilibrium for better representations. Gu et al. [3] proposed a graph learning framework, called Implicit Graph Neural Networks (IGNN), which addresses the finite nature of the underlying recurrent structure struggled by current GNN methods in capturing long-range dependencies in underlying graphs. In this project, we will explore current implicit deep learning models, especially the IGNN models, for fake news early detection.

You are expected to complete COMP9444 and MATH2089 (desirable) to conduct the research in this project.