Excited to announce Relational Foundation Models, moving supervised Relational Deep Learning into the in-context learning setting:. ✅On-the-fly training-free predictions.✅Impressive out-of-the-box performance. Paper: Try For Free:

RT @jure: 🚀 Introducing KumoRFM — the world’s first Relational Foundation Model purpose-built for enterprise prediction tasks!. KumoRFM rea….

RT @Kumo_ai_team: Introducing the new Kumo platform: Build predictive and embedding models directly on your relational data — without featu….

RT @yiwenyuan98: 🎉 Excited to announce the release of ContextGNN, our state-of-art and scalable recommendation model from Kumo!. 🔗 ContextG….

RT @jure: 💠 Stanford Graph Learning Workshop 2024! Join leaders from academia and industry to explore the latest in Machine Learning and AI….

PyG 2.6 is here, including new models and examples on how to combine GNNs with LLMs! Thanks to many contributors who have made this release possible. Full release notes👇.

RT @Kumo_ai_team: What an exciting week it has been at the #KDD conference in Barcelona! We were thrilled to be here and to have two of our….

RT @jure: 🚀 Announcing RelBench: an open benchmark for deep learning on relational databases! RelBench is the foundational infrastructure f….

RT @weihua916: Check out our PyTorch Frame tech report We aim to push tabular deep learning to handle complex mult….

📢 We are hosting a webinar on March 6th 8am PT/5pm CET on the latest PyG release. The webinar covers everything you need to know about PyG 2.5, including a live demo on how set up and use our new distributed GNN training solution. Register for free 👇.

[6/6] As always, PyG brings full support for the latest PyTorch version. Special thanks to all contributors who have made this release possible 🤗

[5/6] PyG 2.5 introduces a full re-implementation of its message passing interface, which makes it natively applicable to torch.compile and TorchScript.

[4/6] Previously, most link prediction models were treated as (non-practical) binary classification tasks. With PyG 2.5, we provide all the tools for using GNNs as a fully-fledged out recommender system, including MIPS and mini-batch-based retrieval metrics.

[3/6] PyG introduces the EdgeIndex class which sub-classes torch.Tensor and extends it by additional metadata. Metadata is maintained and adjusted over its lifespan. This maintains the ease-of-use of regular COO-based workflows, while ensuring optimal message passing computation.

[2/6] PyG 2.5 now fully supports distributed training on large-scale datasets. Includes:.- Balanced graph partitioning via METIS.- DDP for model training, RPC for remote sampling.- Distributed neighbor samplers and loaders. Full tutorial 👇.

📢 Excited to release PyG (PyTorch Geometric) v2.5, including distributed training, a dedicated graph tensor representation, RecSys support, PyTorch 2.2 and native compilation support 🎉. Release Notes 👇. 🧵 A thread [1/6]

RT @weihua916: 🚀🎉 Excited to announce 🌟 PyTorch Frame 🌟 - our new open-source initiative in PyTorch! Dive into multi-modal tabular deep lea….

RT @jure: Curious about what I’ve been working on over the last two years? .Join me next Thursday for the launch of .

[5/5] Since PyG 2.3, we hosted multiple community sprints to extend models, datasets, tutorials and examples. For example, we included Learnable Commutative Monoids, Directed-GNN, PMLP, and many more. Join our Slack for future community sprints 👇.

[4/5] We introduce hierarchical neighbor sampling for maximum performance. This is a trade-off between modular GNN design and efficiency, but using it ensures that we are only using the required amount of information per layer. Full tutorial 👇.