ANR JCJC Funding: ProPruNN

The ProPruNN project aims to use structured pruning in a  hardware-algorithm co-design methodology to improve hardware  implementation of Convolutional Neural Networks (CNNs). The first  sub-hypothesis of the project suggests that designing hardware  architectures that take advantage of structured pruning leads to  significant gains in latency, throughput, and power metrics. However,  this may be complicated in more complex networks like ResNets and  DenseNets, which require rearrangements after filter removal. The second  sub-hypothesis is that it is possible to predict the performance of  pruned networks in terms of accuracy and power, throughput, and latency  metrics. The final sub-hypothesis is that prediction models can be used  to design state-of-the-art networks and push the Pareto frontier of  accuracy vs implementation performances of current literature. The project aims to improve the results of this method by  using a finer and more efficient pruning approach.