Fig. 3
From: Accurate and efficient floor localization with scalable spiking graph neural networks
System architecture of FloorLocator. The input of FloorLocator is a WiFi fingerprint, and its output is the estimated floor label for the given fingerprint. It consists of one TAGConv layer, two DeepBlocks, one transition layer (which is also a TAGConv layer), two FC layers and one voting layer. Each DeepBlock is composed of three DeepLayers, and these DeepLayers are densely connected. Note that each TAGConv layer is followed by an LIF activation and event-based batch normalization layer, while each FC layer is followed only by an LIF activation. These subsequent layers are not shown in the figure for clarity