PinnedPublished inTDS ArchiveThe Road to Biology 2.0 Will Pass Through Black-Box DataFuture bio-AI breakthroughs will arise from novel high-throughput low-cost AI-specific “black-box” data modalities.Mar 18, 20246492Mar 18, 20246492
Published inTDS ArchiveCo-operative Graph Neural NetworksA new message-passing paradigm where every node can choose to either ‘listen’, ‘broadcast’, ‘listen & broadcast’ or ‘isolate’.Dec 6, 20232632Dec 6, 20232632
Published inTDS ArchiveTopological Generalisation with Advective Diffusion TransformersA new diffusion-based continuous GNN model offers better generalisation capabilitiesOct 19, 20232131Oct 19, 20232131
Published inTDS ArchiveDynamically rewired delayed message passing GNNsDynamic rewiring and delayed message passing mechanisms offer a tradeoff between standard MPNNs and graph TransformersJun 19, 2023115Jun 19, 2023115
Published inTDS ArchiveDirection Improves Graph LearningHow a wise use of direction when doing message passing on heterophilic graphs can result in very significant gains.Jun 8, 20234574Jun 8, 20234574
Published inTDS ArchiveHyperbolic Deep Reinforcement LearningMany RL problems have hierarchical tree-like nature. Hyperbolic geometry offers a powerful prior for such problems.Apr 30, 20235314Apr 30, 20235314
Published inTDS ArchiveLearning Network GamesHow to learn the network underlying the interactions of players in social applications, economics, and beyond.Apr 20, 2023269Apr 20, 2023269
Published inTDS ArchiveGraph Neural Networks as gradient flowsGNNs derived as gradient flows minimising a learnable energy that describes attractive and repulsive forces between graph nodes.Oct 14, 20225962Oct 14, 20225962
Published inTDS ArchiveTowards Geometric Deep Learning IV: Chemical Precursors of GNNsIn the last post in the “Towards Geometric Deep Learning” series, we look at early prototypes of GNNs in the field of chemistry.Jul 25, 20225742Jul 25, 20225742
Published inTDS ArchiveTowards Geometric Deep Learning III: First Geometric ArchitecturesIn the third post of our series “Towards Geometric Deep Learning” we look at the first “geometric” architectures: Neocognitron and CNNsJul 18, 2022583Jul 18, 2022583
Published inTDS ArchiveTowards Geometric Deep Learning II: The Perceptron AffairIn the second post of our series “Towards Geometric Deep Learning” we discuss how the criticism of Perceptrons led to geometric insightsJul 11, 2022Jul 11, 2022
Published inTDS ArchiveTowards Geometric Deep Learning I: On the Shoulders of GiantsIn a new series of posts, we discuss how geometric ideas of symmetry underpinning Geometric Deep Learning have emerged through history.Jul 4, 2022Jul 4, 2022
Published inTDS ArchiveAccelerating and scaling Temporal Graph Networks on the Graphcore IPUIs GPU the best hardware choice for GNNs? Together with Graphcore, we explore the advantages of the new IPU architecture for temporal GNNs.Jun 14, 20223Jun 14, 20223
Published inTDS ArchiveA new computational fabric for Graph Neural NetworksAre graphs the right computational fabric for GNNs? A recent line of papers challenges this assumption.Jun 10, 20223Jun 10, 20223
Published inTDS ArchiveNeural Sheaf Diffusion for deep learning on graphsCellular sheaf theory, a branch of algebraic topology, provides new insights into how Graph Neural Networks work and how to design new…May 16, 2022May 16, 2022
Published inTDS ArchiveAnnouncing the Learning on Graphs ConferenceGraph Machine Learning has become large enough of a field to deserve its own standalone event: the Learning on Graphs Conference (LoG).Apr 15, 20221Apr 15, 20221
Published inTDS ArchiveGraph Neural Networks beyond Weisfeiler-Lehman and vanilla Message PassingPhysics-inspired continuous learning models on graphs allow to overcome the limitations of traditional GNNsMar 3, 20224Mar 3, 20224
Published inTDS ArchiveLearning on graphs with missing featuresFeature Propagation is a simple and surprisingly efficient solution for learning on graphs with missing node featuresFeb 3, 20223Feb 3, 20223
Published inTDS ArchivePredictions and hopes for Geometric & Graph ML in 2022Leading researchers in Geometric & Graph ML summarise the progress in 2021 and make predictions for 2022Jan 24, 20225Jan 24, 20225
Published inTDS ArchiveUsing subgraphs for more expressive GNNsGNNs have limited expressive power due to their equivalence to WL test. Recent works show how to improve expressivity by using subgraphs.Dec 20, 20211Dec 20, 20211