Spark TFOCS Optimization Package

Agenda:

6:30-7:00pm: Mingling

7:00-7:05pm: Intro's

7:05-8:15pm: Technical Talk

8:15-8:30: Mingling

This meetup will be filmed and uploaded to the Apache Spark YouTube page.

Abstract:

The spark-tfocs package ports TFOCS to Spark. TFOCS is a Matlab library providing templates for first-order conic solvers. In this talk we will first provide an overview of the convex optimization methods used in TFOCS, including Nesterov’s accelerated gradient descent. We will then discuss the interfaces used to define optimization problems in spark-tfocs, using examples such as LASSO, linear programming, and quantile regression. We will discuss some of the design and implementation choices made in the development and then demonstrate solution of the example problems.

Bio:

Aaron Staple is the author of the spark-tfocs package, with support from Databricks. Aaron is a freelance developer with experience in data infrastructure and analytics. Aaron was previously an early employee at MongoDB, where he implemented the MongoDB query engine.
Before that he wrote software for a distributed video streaming server designed by Andy Bechtolsheim and David Cheriton. Aaron studied Math and Computer Science at Stanford University.