Google Code-in

Flux.jl has a lot of models predefined, including many of the commonly used ones including ResNet and ImageNet etc. The benchmarks for these operations is a little out of date, and needs some attention. It is important to have up to date benchmarks, especially in light of a lot of the changes going around in the ecosystem.

You would be required to benchmark some of the common tasks that come up frequently in machine learning, like convolutions, taking gradients of some common functions etc which we would like to compare against Tensorflow and PyTorch. Functions to be benchmarked : Conv, DepthwiseConv, ConvTranspose, Dense, LSTM, RNN, Normalization layers, and CrossCor.