Fig. 3: Classification of collaboration networks.
From: Echo state graph neural networks with analogue random resistive memory arrays
a, Example collaboration network graphs from the COLLAB dataset that correspond to different branches of physics: astrophysics (AP), high energy physics (HE) and condensed matter physics (CM). Each node denotes a researcher, while an edge represents a collaboration relation. b, An example COLLAB node embedding process according to the protocol shown in Fig. 1f and Methods, which leads to node embeddings that encapsulate more graph information. c, Graph embedding vectors of the three categories of the COLLAB dataset. Each column is a graph embedding. d, Graph embeddings mapped to a 2D space using PCA. Orange, blue and purple dots denote collaboration networks from the AP, CM and HE communities, respectively, revealing a clear boundary between AP and CM. e, The accuracy of each fold in a ten-fold cross-validation and the software baseline. The average accuracy is 73.00%, comparable to state-of-the-art algorithms. f, The confusion matrices of the experimental classification results. The upper matrix is a ten-fold averaged confusion matrix, which is then normalized horizontally to produce the lower matrix. g, A breakdown of the estimated OPs (red bars) and associated energy (light-blue bars for a state-of-the-art GPU; dark-blue bars for a projected random resistive memory-based hybrid analogue–digital system). In a forward (backward) pass, the fully optimized model on a state-of-the-art GPU and ESGNN on a projected random resistive memory-based hybrid analogue–digital system consume approximately 8.31 mJ (approximately 16.98 GOPs) and approximately 234.59 μJ (approximately 1.16 MOPs), respectively, revealing a >35.42 fold improvement of the inference energy efficiency (an approximately 99.99% reduction of the backward pass complexity).
