A Picture is Worth a Billion Bits: Adaptive Visualization of Big Data
Abstract
This report documents the results achieved by the University of Washington SandCat team under the Defense Advanced Research Projects Agency Building Resource Adaptive Software Systems program. This effort focused on developing static and dynamic adaptations of computer systems, as a modern system stack is forced to (1) accommodate the growth in data volume and data bandwidth at rates exceeding Moore's Law; and (2) support energy-efficient mobile devices that can neither store all necessary data nor perform the computation at the desired real-time latency. To meet these two goals, SandCat technology automates adaptation by aggressively approximating both the data and the computation, as well as by performing the computations incrementally. In particular, the Sandcat project responded to two needs for adaptation: (1) the rapid change in hardware, systems software, and application workloads; and (2) the increased complexity of computer systems, especially at the interfaces, which were intended to hide the complexity. A major milestone in the research area of program synthesis was achieved, as the synthesis technology developed by SandCat reached parity with human programmers on about a dozen different programming tasks. Several components of SandCat technology have been transitioned into the commercial marketplace, including verified lifting which ships in a commercial photo editing product; the Rosette architecture which is being used in the Synthetic Minds startup; and the machine learning kernels which are being commercialized by OctoML.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2020
- Accession Number
- AD1104597
Entities
People
- Alvin Cheung
- Dan Grossman
- Emina Torlak
- Luis Ceze
- Michael Ernst
- Rastislav BodÃk
- Xi Wang
- Zachary Tatlock
Organizations
- University of Washington