English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Many real world graphs contain time domain information. Temporal Graph NeuralNetworks capture temporal information as well as structural and contextualinformation in the generated dynamic node embeddings. Researchers have shownthat these embeddings achieve state-of-the-art performance in many differenttasks. In this work, we propose TGL, a unified framework for large-scaleoffline Temporal Graph Neural Network training where users can compose variousTemporal Graph Neural Networks with simple configuration files. TGL comprisesfive main components, a temporal sampler, a mailbox, a node memory module, amemory updater, and a message passing engine. We design a Temporal-CSR datastructure and a parallel sampler to efficiently sample temporal neighbors toformtraining mini-batches. We propose a novel random chunk scheduling techniquethat mitigates the problem of obsolete node memory when training with a largebatch size. To address the limitations of current TGNNs only being evaluated onsmall-scale datasets, we introduce two large-scale real-world datasets with 0.2and 1.3 billion temporal edges. We evaluate the performance of TGL on foursmall-scale datasets with a single GPU and the two large datasets with multipleGPUs for both link prediction and node classification tasks. We compare TGLwith the open-sourced code of five methods and show that TGL achieves similaror better accuracy with an average of 13x speedup. Our temporal parallelsampler achieves an average of 173x speedup on a multi-core CPU compared withthe baselines. On a 4-GPU machine, TGL can train one epoch of more than onebillion temporal edges within 1-10 hours. To the best of our knowledge, this isthe first work that proposes a general framework for large-scale Temporal GraphNeural Networks training on multiple GPUs.

TGL: A General Framework for Temporal GNN Training on Billion-Scale Graphs