Performance analysis tools for parallel applications have been studied for decades. The current convergence between HPC and IA changes the software stack and require a redesign of the performance analysis tools.
The goal of this project is to develop new performance analysis techniques that target machine learning frameworks such as tensorflow.