Machine Learning and Networks - Challenges, Solutions and Tradeoffs

Abstract

We consider the intersection of networks and  machine learning in two  contexts.   In the first, the data of interest for ML  is in the form of a  large,  complex network.  Here, Graph Neural Networks (GNNs) utilize message passing for neighborhood aggregation to capture graph topology, while Graph Embedding based Neural Networks (GENNs) distill essential graph information into a concise representation suitable for traditional neural architectures.  The second is when the training and/or  the  evaluation phase of ML must be carried out over a distributed environment such as an IoT network. These environments pose challenges due to limited storage, communication and power, complicating the deployment of complex ML models and impeding real-time decision-making. Our innovations in  graph-coordinate based strategies, TCNN and DVCNN, help sidestep the computational challenges faced by competing algorithms. Experimental results, benchmarked against the Open Graph Benchmark Leaderboard, demonstrate that TCNN and DVCNN require orders of magnitude fewer parameters than any neural network method currently listed in the OGBN Leaderboard for both OGBNProteins and OGBN-Products datasets.  

About this Lecture

Request this Lecture

To request this particular lecture, please complete this online form.

Request a Tour

To request a tour with this speaker, please complete this online form.

All requests will be sent to ACM headquarters for review.