Fig. 4: Comparison of the performance of EGC in various applications with existing models.

a Results of pathfinding (upper panel) and on-condition pathfinding (lower panel) using EGC and LM algorithm. In the upper panel, each symbol represents the average number of attempts, and the error bars denote the standard deviation across trials. b, c Results of link prediction and community detection using EGC and software-based algorithms. d EGC-based PPI prediction and PageRank algorithm. In the PPI network-mapped CBA, a 3-hop similarity for the PPI prediction can be achieved by deactivating the common proteins between two proteins (upper panel). The PageRank algorithm can be implemented with steady-state estimation of the EGC-based probabilistic graph. e, f Comparison of energy consumption and latency between CPU-based graph embedding and memristive PGR. The energy consumption and latency of the CPU-based graph embedding were calculated based on a single core of the Intel Core i7-10700K. The memristive PGR results were estimated based on the power and latency of the devices and peripheral circuits involved in physically mapping the graph. The graph density was fixed at 0.1 for all comparisons.