Sandeep Agrawal

I currently work for Oracle Labs in Automated Machine Learning (AutoML). I am primarily responsible for Automatic Algorithm Selection, which aims to identify the best algorithm for a given dataset and learning task, significantly reducing the time to deploy Machine Learning applications. We achieve by using meta-learning, a machine learning technique that aims to learn algorithm performance from prior workloads, and uses this knowledge to infer algorithm performance for a new workload.

Prior to AutoML, I was a part of the RAPID project, where I was responsible for the scalability of Machine Learning algorithms across a massive number of nodes. By carefully interleaving compute and network, we were able to scale-out the Sparse Matrix-Matrix Multiplication algorithm used in Similarity search algorithm to more than 256 nodes! You can find publications related to my RAPID work here - https://dl.acm.org/citation.cfm?id=3123985 (MICRO 2017), https://dl.acm.org/citation.cfm?id=3190655 (SIGMOD 2018) and https://ieeexplore.ieee.org/abstract/document/8029318/ (BigData Congress 2017).

I completed my PhD in Computer Science from Duke University, under the guidance of Prof. Alvin Lebeck. My thesis primarily focused on data parallel hardware, and how it achieves high energy efficiency by amortizing instruction costs across multiple data streams, and high throughput by enabling massive parallelism across independent threads. You can take a look at my thesis here - https://dukespace.lib.duke.edu/dspace/handle/10161/9957. Relevant publications are https://dl.acm.org/citation.cfm?id=2541956 (ASPLOS 2014) and https://dl.acm.org/citation.cfm?id=2851144 (PPoPP 2016).

I also got my Masters degree in the Dept. of Electrical and Computer Engineering here at Duke. I spent the summer of 2011 working at Swype in Seattle, and both the company and city are experiences I will cherish forever.

Prior to Duke I spent an amazing four years at Samsung India working on their native Linux platform. My work at Samsung was primarily aimed towards graphics optimizations for the various chipsets the platform ran on. As a part of the UI framework team, I also spent the better part of 2 years working at Samsung South Korea hacking away at the entire graphics stack from GTK to the EGL infrastructure to the 2D raster backend in pixman.

I completed my Bachelors in Electronics and Communication Technology at the Indian Institute of Technology (IIT) Roorkee, a beautiful city surrounded by the Himalayas.