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 Plus annual

Presents the access of premium content as premium feature

Premium Content

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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

As graph data size increases, the vast latency and memory consumption duringinference pose a significant challenge to the real-world deployment of GraphNeural Networks (GNNs). While quantization is a powerful approach to reducingGNNs complexity, most previous works on GNNs quantization fail to exploit theunique characteristics of GNNs, suffering from severe accuracy degradation.Through an in-depth analysis of the topology of GNNs, we observe that thetopology of the graph leads to significant differences between nodes, and mostof the nodes in a graph appear to have a small aggregation value. Motivated bythis, in this paper, we propose the Aggregation-Aware mixed-precisionQuantization ($\rm A^2Q$) for GNNs, where an appropriate bitwidth isautomatically learned and assigned to each node in the graph. To mitigate thevanishing gradient problem caused by sparse connections between nodes, wepropose a Local Gradient method to serve the quantization error of the nodefeatures as the supervision during training. We also develop a Nearest NeighborStrategy to deal with the generalization on unseen graphs. Extensiveexperiments on eight public node-level and graph-level datasets demonstrate thegenerality and robustness of our proposed method. Compared to the FP32 models,our method can achieve up to a 18.6x (i.e., 1.70bit) compression ratio withnegligible accuracy degradation. Morever, compared to the state-of-the-artquantization method, our method can achieve up to 11.4\% and 9.5\% accuracyimprovements on the node-level and graph-level tasks, respectively, and up to2x speedup on a dedicated hardware accelerator.

$\rm A^2Q$: Aggregation-Aware Quantization for Graph Neural Networks