Dr Petar Veličković

blankphoto.png

Speaker

DeepMind

Petar Veličković is a Senior Research Scientist at DeepMind. He holds a PhD in Computer Science from the University of Cambridge (Trinity College), obtained under the supervision of Pietro Liò. His research interests involve devising neural network architectures that operate on nontrivially structured data (such as graphs), and their applications in algorithmic reasoning and computational biology. He has published relevant research in these areas at both machine learning venues (NeurIPS, ICLR, ICML-W) and biomedical venues and journals (Bioinformatics, PLOS One, JCB, PervasiveHealth). In particular, he is the first author of Graph Attention Networks—a popular convolutional layer for graphs—and Deep Graph Infomax—a scalable local/global unsupervised learning pipeline for graphs (featured in ZDNet). Further, his research has been used in substantially improving the travel-time predictions in Google Maps (covered by outlets including the CNBC, Endgadget, VentureBeat, CNET, the Verge and ZDNet).

Talk

Type-checking graph neural networks

Recent advances in neural algorithmic reasoning with graph neural networks (GNNs) are propped up by the notion of algorithmic alignment. Broadly, a neural network will be better at learning to execute a reasoning task (in terms of sample complexity) if its individual components align well with the target algorithm. Specifically, GNNs are claimed to align with dynamic programming (DP), a general problem-solving strategy which expresses many polynomial-time algorithms. However, has this alignment truly been demonstrated and theoretically quantified? Here we show, using methods from category theory and abstract algebra, that there exists an intricate connection between GNNs and DP, going well beyond the initial observations over individual algorithms such as Bellman-Ford. Exposing this connection, we easily verify several prior findings in the literature. But more generally, our approach allowed us to immediately detect a misalignment of previous proposals for using GNN architectures for edge-centric tasks, and propose a fixed implementation, in a manner not at all unlike type checking. Our proposal demonstrates strong empirical results on the CLRS algorithmic reasoning benchmark, and we hope our exposition will serve as a foundation for building stronger algorithmically aligned GNNs in the future.