[Colloquium] Bruno Barbarioli Dissertation Defense/Sep 22, 2023
meganwoodward at uchicago.edu
meganwoodward at uchicago.edu
Mon Sep 11 09:50:50 CDT 2023
This is an announcement of Bruno Barbarioli's Dissertation Defense.
===============================================
Candidate: Bruno Barbarioli
Date: Friday, September 22, 2023
Time: 1 pm CST
Location: JCL 346
Title: A Principled Framework for Adaptive Lossy Data Compression with Linfinity Error Guarantees
Abstract: Recent advances in edge computing and networking have led to a renaissance in the field of ubiquitous sensing. Emerging applications range from imaging in autonomous vehicles to geospatial data in remote satellite imagery. For such data, storage and communication is a major bottleneck and lossy data compression can help. However, lossy data compression is not as well developed beyond "perceptual" data formats such as images, audio, and video. The errors such algorithms introduce can interact with downstream sensing algorithms and lead to unforeseen consequences. This dissertation develops a principled approach to lossy data compression over structured arrays of floating-point data. It presents algorithms that guarantee the $l_\infty$ error, or the maximum error over all elements. These algorithms rely on different combinations of floating point quantization and spatial quantization, both of which preserve $l_\infty$ guarantees, and thus can be adaptively interspersed depending on the data. Results show improved compression ratios for high-resolution geospatial raster data and novel algorithms for optimizing edge-resident time-series data.
Advisors: Sanjay Krishnan
Committee Members: Aaron Elmore, Sanjay Krishnan, Raul Castro Fernandez, and Stavros Sintos
More information about the Colloquium
mailing list