Balance-aware Sequence Sampling Makes Multi-modal Learning Better
Zhi-Hao Guan · 2025
Read Balance-aware Sequence Sampling Makes Multi-modal Learning Better by Zhi-Hao Guan (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-2501-01470.txt.
Description
Zhi-Hao Guan, writing solo in 2025, argues that the order of training examples can fix lopsided multi-modal learning. When a model learns from mixed inputs like text and images, one mode often races ahead while the other lags, and most fixes tweak the training goal.
Guan points at a neglected lever: the sequence in which samples arrive. A measurer scores how balanced each sample is across modes. A scheduler then feeds the model balanced samples first and harder, lopsided ones later, following curriculum learning. A second method reshuffles the odds each round as the model grows stronger. Tests on standard data sets beat leading rivals, showing order alone can rebalance learning.
Catalog record
| CREATOR | Zhi-Hao Guan |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-01470