This repository has been archived and is no longer maintained. Please visit the new repository at BrainchipInc/Pleiades.

Paper: Building Temporal Kernels with Orthogonal Polynomials

This is a self-contained repo for using temporal kernels parameterized by orthogonal polynomials. For example, it can be used as a drop-in replacement for convolutional layers (only supporting nn.Conv3d layers for now), where the last dimension (assumed to be temporal) will be parameterized by orthogonal polynomials up to a given degree.

from model import PleiadesLayer

layer = PleiadesLayer(2, 8, kernel_size=(3, 3, 20), degrees=4)

The structured temporal kernels can also easily be resampled into different kernel sizes without needing to retrain the network.

layer.resample(10)  # downsample the kernel size from 20 to 10