Riyadh Baghdadi is a postdoctoral researcher at CSAIL and received his PhD from INRIA/Sorbonne UPMC in Paris. Baghdadi worked in the area of polyhedral compilation, which uses a mathematical model to allow compilers perform advanced code optimizations. This area has been the foundation of his work throughout his postdoc.
Baghdadi’s research interests include the intersection of applied machine learning and compilers, as well as compilers and programming models for high performance computing and compute intensive areas.
“I’m excited by two things: making things run faster and developing intelligent systems,” Baghdadi said. “The intersection of these two areas gives birth to my area of research!”
Baghdadi said he’s always been excited about the amount of innovation at CSAIL. He is part of the Commit research team advised by Professor Saman Amarsinghe. The team’s goal is to develop new programming languages and compilers in order to generate efficient code for different hardware architectures.
Baghdadi said deep learning models are becoming more computationally expensive.
“For example, if you want to train a state of the art deep learning model, you need a very large cluster of computers of GPUs which are very expensive,” he said. “Also, it is hard to deploy these neural networks on smartphones or embedded systems with limited resources, such as driverless cars.”
The goal is to develop compilers that can take these models and make them run faster so that you can do the training on a smaller number of resources.
Baghdadi has led the team that developed Tiramisu, a polyhedral compiler for deep learning, tensor algebra and image processing. It is the first polyhedral compiler to generate code that matches or outperforms highly optimized industrial linear algebra and deep learning libraries. Compared to state-of-the-art DNN compilers, Tiramisu has the unique ability to parallelize multi-layer RNNs and to exploit DNN weight sparsity. It is open source and has a growing community of contributors from more than 10 industrial and academic institutions.
One of the things Baghdadi most appreciates about working at CSAIL is the spirit and culture of tackling new and difficult challenges.
“I think that makes [CSAIL] very interesting, and also makes the culture very unique,” Baghdadi said.
Following his time at CSAIL, Baghdadi will work as an assistant professor at New York University Abu Dhabi.