Balancing Information Preservation and Disentanglement in Self-Supervised Music Representation Learning
2025
Read Balancing Information Preservation and Disentanglement in Self-Supervised Music Representation Learning (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-22995.txt.
Description
Julia Wilkins and co-authors teach computers to split music into separate traits in a 2025 paper. Self-taught music models face a tug of war: rebuild the sound faithfully and you keep every detail, but contrast similar clips and you gain clean meaning. Few studies test both drives in one frame.
Their multi-view setup blends the two goals so that distinct traits, like instrument or mood factors, land in separate parts of the learned code. Careful tests of design choices show each goal alone trades off against the other. Combined well, they disentangle music traits without losing information, the authors report.
Catalog record
| YEAR | 2025 |
|---|---|
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-22995