
On Local Aggregation in Heterophilic Graphs
Many recent works have studied the performance of Graph Neural Networks ...
read it

Node Embedding using Mutual Information and SelfSupervision based Bilevel Aggregation
Graph Neural Networks (GNNs) learn low dimensional representations of no...
read it

Destructive Read by Wave Interference for Arbitration
With the advent of big data and deep learning, computation power has bec...
read it

Learning the geometry of wavebased imaging
We propose a general deep learning architecture for wavebased imaging p...
read it

Are Graph Convolutional Networks Fully Exploiting Graph Structure?
Graph Convolutional Networks (GCNs) generalize the idea of deep convolut...
read it

Graphbased Pyramid Global Context Reasoning with a Saliencyaware Projection for COVID19 Lung Infections Segmentation
Coronavirus Disease 2019 (COVID19) has rapidly spread in 2020, emerging...
read it

Graph Neural Networks Inspired by Classical Iterative Algorithms
Despite the recent success of graph neural networks (GNN), common archit...
read it
Deep learning longrange information in undirected graphs with wave networks
Graph algorithms are key tools in many fields of science and technology. Some of these algorithms depend on propagating information between distant nodes in a graph. Recently, there have been a number of deep learning architectures proposed to learn on undirected graphs. However, most of these architectures aggregate information in the local neighborhood of a node, and therefore they may not be capable of efficiently propagating longrange information. To solve this problem we examine a recently proposed architecture, wave, which propagates information back and forth across an undirected graph in waves of nonlinear computation. We compare wave to graph convolution, an architecture based on local aggregation, and find that wave learns three different graphbased tasks with greater efficiency and accuracy. These three tasks include (1) labeling a path connecting two nodes in a graph, (2) solving a maze presented as an image, and (3) computing voltages in a circuit. These tasks range from trivial to very difficult, but wave can extrapolate from small training examples to much larger testing examples. These results show that wave may be able to efficiently solve a wide range of problems that require longrange information propagation across undirected graphs. An implementation of the wave network, and example code for the maze problem are included in the tflon deep learning toolkit (https://bitbucket.org/mkmatlock/tflon).
READ FULL TEXT
Comments
There are no comments yet.