Python

Improving mapping of computational tasks to manycore architectures using computational group theory, published in IEEE TCAD. MPsym can determine whether different mappings are equivalent due to symmetries of an architecture’s topology. Works efficiently with large, hierarchical architectures and can be used to improve existing mapping discovery algorithms by making them “symmetry aware”. Implemented in C++ for performance with easy to use Python bindings.

A from-scratch PyTorch implementation of the convolutional neural network described by Zhang et al. in Colorful Image Colorization. Can be used with pretrained weights or trained from scratch. Also includes an implementation of Google’s DeepLab-V3+ network that can be substituted for the default VGG network.