Vivek Nair


I am a Research Scientist at Facebook in Seattle. I received my Ph.D. in Computer Science from North Carolina State University in 2018, where Dr. Tim Menzies was my advisor. I was previously a Reseach Intern at Microsoft Research, and Lexisnexis Risk Solutions. Before that a Software Engineer at Samsung Research Institute - Noida working in Memory & File System.


Performance Prediction of Configurable System: As a part of my Ph.D., I explored ways to easily optimize non-functional properties (e.g. runtime, latency etc.) of highly configurable systems. Techniques like spectral learning and rank based learning is used to learn features of the configurations.

Finding Best Cloud Configuration for a given Software System: This is a joint project with my colleague, Chin-Jung Hsu. We wanted to develop fast searching techniques, which are efficient, effective and robust, to find the best configuration in the cloud. The methods proposed uses low-level performance metrics to predict resource requirements which can be then used to eliminate 'less promising' configurations and hence searches only the spotlight region of the configuration space.

SWAY - Sampling Way: This is a joint project with my colleague, Jianfeng Chen. We wanted to develop fast searching techniques, which can be used to optimize problems rather quickly. We have tried the searchers on various problems. This has resulted in few publications.


Open Source Contributions

HPCC Spark Integrator: At LexisNexis, I worked on the developing a python package, which can be used to import (sampled) data stored in HPCC cluster to a local machine. This was achieved using the web services exposed by the HPCC platform. The prototype is available here.

HPCC - Data Science Portal - Machine Learning Plugins: During Summer'16, I worked on developing machine learning plugins for Data Science Portal (DSP). DSP is a graphical toolkit, which runs HPCC under the hood. It relieves the users to learn or optimize ECL code and provides a graphical interface to generate machine learning pipelines. The code required to adapt the code to DSP required careful porting of the existing ML library. The pull request is available here. The work was presented at the HPCC summit as a part of the HPCC poster competition. The copy of the poster is available here.

HPCC - Testing Suite: Since, HPCC ML suite was in its nascent stage, I had an opportunity to develop the testing suite for the ML suite. It was particularly a very interesting because I got to learn how to use the macros and functionmacros, which are indeed very powerful. The pull request can be found here.




A naive stab at extending a framework, JMOO - originally developed by Joe Krall . The newer and extended version is called Storm. The newer version has new algorithms like MOEA/D, NSGA-III along with various other performance measures.