fedepare / cuSNN-samples

Samples which demonstrate features provided by cuSNN

cuSNN-samples

Samples for cuSNN developers which demonstrate the main features provided by the library.

At the moment, all samples available are to the motion-selective hierarchical SNN proposed in "Unsupervised Learning of a Hierarchical Spiking Neural Network for Optical Flow Estimation: From Events to Global Motion Perception" (Paredes-Vallés, F., Scheper, K.Y., and de Croon, G.C.H.E., 2018). More varied samples will be included in the near future.

Samples list

• train-SSConv: Train a sigle-synaptic Conv2d (SS-Conv) layer.
• test-SSConv: Test an SS-Conv layer.
• train-MSConv: Train a multi-synaptic Conv2d (MS-Conv) layer, preceded by a pre-trained SS-Conv layer.
• test-MSConv: Test a three-layer network with a SS-Conv, a Merge, and an MS-Conv layer.
• train-Dense: Train a Dense layer, preceded by pre-trained SS-Conv and MS-Conv layers.
• test-MSConv: Test a five-layer network with a SS-Conv, a Merge, an MS-Conv, a Pooling, and a Dense layer.