Getting most out of representations - Approaches to reducing information redundancy
Date
Friday, September 9, 2022 10:00 - 11:00
Speaker
Karen Ullrich (FAIR NY & Vector Institute)
Location
Heinzel Seminar Room / Office Bldg West (I21.EG.101)
Series
Seminar/Talk
Tags
ELLIS talk
Host
Christoph Lampert
Contact
I will talk about algorithms that find and eliminate redundant information in representations of data. The solutions I present are heavily inspired by classical information theory and modern machine learning. Relevant papers are:
[1] Compressing multisets with large alphabets
D Severo, J Townsend, A Khisti, A Makhzani, K Ullrich - 2022 Data Compression Conference (DCC), 2022
[2] Lossy compression for lossless prediction
Y Dubois, B Bloem-Reddy, K Ullrich, CJ Maddison - Advances in Neural Information Processing Systems, 2021