Anders Host-Madsen
University of Hawaii at Manoa, HI·Engineering
Every day the world produces hundreds of thousands of billions of bytes of video, images, music and scientific data. Storing and sending this data, often wirelessly, depends on compression, which removes redundancy so that fewer bits describe the same content.
- Fields
Engineering
Computer Science
Mathematics & Statistics
- Topics
data compression
source coding
lossless compression
- Grants
- 1 known grant, through Sep 2027
What they work on
- Derive fundamental performance bounds for machine learning of lossless source coding within the probably approximately correct (PAC) learning framework, and identify which factors control how fast learning converges.
- Develop active learning methods for compression that decide how much data to examine, using stopping rules to detect when a simple underlying model is sufficient.
- Determine in what sense lossy coding, including almost-lossless coding and lossless coding of real-valued sources, can be learned under uniform and non-uniform PAC frameworks, and derive performance bounds, particularly for sources with memory.
More about the research
Some compression is lossy, like JPEG photos, which lose a little quality. Some is lossless, like Zip files, which keep every bit. Machine learning has started to beat traditional compression methods, but there is little theory explaining when and why it works.
Why it matters
Better theory for machine-learning compression could lead to more powerful compression methods for video, images, music and scientific data. This could make more efficient use of wireless spectrum and reduce energy use in mobile devices for applications such as streaming, video calls and telemedicine.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Anthony Sicilia
West Virginia University, WV·Computer Science
Develop a Probably Approximately Correct (PAC) learning theory that measures how user disagreement, suggestions and topic shifts change the calibration of an AI system's confidence during a conversation.
Hessam Mahdavifar
Northeastern University, MA·Engineering·Early-career
Generalize Plotkin Transform codes, the family containing Reed–Muller and polar codes, through nonlinear parameterizations, and search for new codes using machine learning, deep learning and gradient-descent optimization.
Paul Valiant
Purdue University, IN·Computer Science
Design error-correcting codes that withstand random or adversarial insertion and deletion errors while supporting local decoding, which recovers part of a message by reading only a small piece of the stored data.
The school
University of Hawaii at Manoa: 23 labs on PI Pages, $376M a year on research, 182 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.