Frameworks like Spark and TensorFlow have commoditized cluster computing and training of neural networks. However, they leave precious performance on the table, especially when used together. Flare is a new back-end for Spark SQL that yields significant speedups by compiling query plans to native code. Lantern takes a fresh look at backpropagation, and provides a generic and performant differentiable programming framework in Scala, including efficient code generation for GPUs. This talk discusses interesting design aspects of Flare and Lantern and presents relevant case studies.
Basic Scala. Experience with Spark and/or Deep Learning a plus.
Tiark Rompf is an Assistant Professor at Purdue University. His work focuses on advanced compiler technology and associated language support. From 2008 to 2014 he was a member of the Scala team at EPFL, where he contributed to the Scala language and toolchain (continuations, data structures, compiler performance, type system proofs). From 2012 to 2014 he was a researcher at Oracle Labs. His work received a number of awards (NSF CAREER, VMware Systems Research Award, several Best Paper/Artifact).