Debloating Machine Learning Systems

Updated: almost 3 years ago
Deadline: 08 Oct 2021

Machine learning systems are proliferating our life, being part of almost every modern software stack. From industrial automation, to healthcare systems, there is an exponential adoption of these systems. While a lot of research is being done on how to design and build machine learning algorithms, very little work is being done on optimizing the systems that run these algorithms.

Like many computer systems, machine learning systems are bloated, with millions of lines of code being added to these systems every year. This bloat results in a technical debt when using these systems, where some resources (and energy) might be wasted due to this bloat. Besides the wasted resources, in general, code bloat leads to increased security vulnerabilities.

This project will investigate how to optimize machine learning systems, building on some of our recent results on removing code bloat where it exists. In addition, we will look at how to increase the modularity of these systems, to be more manageable, scalable, and efficient.



Similar Positions