Publications

. InferLine: ML Inference Pipeline Composition. 2018.

Preprint

. Research for Practice: Prediction-Serving Systems . In ACM Queue, 2018.

URL

. IDK Cascades: Fast Deep Learning by Learning not to Overthink. 2017.

Preprint

. Clipper: A Low-Latency Online Prediction Serving System. In NSDI, 2017.

PDF Code Project Slides Video

. Scalable Training and Serving of Personalized Models. In LearningSys, 2015.

PDF

. The Missing Piece in Complex Analytics: Low Latency, Scalable Model Management and Serving with Velox. In CIDR, 2015.

PDF Code Project

. GraphX: Graph Processing in a Distributed Dataflow Framework. In OSDI, 2014.

PDF Project Slides

. Inverted Indices for Particle Tracking in Petascale Cosmological Simulations. In SSDBM, 2013.

PDF

Projects

Clipper

Clipper is general-purpose, low-latency prediction serving system.

Flor

Flor is a system for building, configuring, and tracking machine-learning workflows.

GraphX

GraphX is Apache Spark’s API for graphs and graph-parallel computation.

Velox

Velox is a system for serving machine learning predictions, featuring tight integration with Apache Spark and KeystoneML, model retraining that spans online and batch learning, and automatic support for personalized predictions.