Petar Veličković

I hold an Affiliated Lectureship at the University of Cambridge, where I am co-teaching an MPhil course on geometric deep learning, and I am an Associate of Clare Hall, Cambridge. I am currently employed as a Staff Research Scientist at Google DeepMind. I hold a PhD in Computer Science from the University of Cambridge (Trinity College), obtained under the supervision of Pietro Liò.

My research concerns geometric deep learning—devising neural network architectures that respect the invariances and symmetries in data (a topic I’ve co-written a proto-book about). For my contributions, I am recognised as an ELLIS Scholar in the Geometric Deep Learning Program. Particularly, I focus on graph representation learning and its applications in algorithmic reasoning (featured in VentureBeat).

I am the first author of Graph Attention Networks—a popular convolutional layer for graphs—and Deep Graph Infomax—a popular self-supervised learning pipeline for graphs (featured in ZDNet).

My 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), and guiding intuition of mathematicians towards new top-tier theorems and conjectures (featured in Nature, Science, Quanta Magazine, New Scientist, The Independent, Sky News, The Sunday Times, la Repubblica and The Conversation).