Weight Factorization and Centralization for Continual Learning in Speech Recognition
Ugan, Enes Yavuz, Pham, Ngoc-Quan, Waibel, Alexander
–arXiv.org Artificial Intelligence
Modern neural network based speech recognition models are required to continually absorb new data without re-training the whole system, especially in downstream applications using foundation models, having no access to the original training data. Continually training the models in a rehearsal-free, multilingual, and language agnostic condition, likely leads to catastrophic forgetting, when a seemingly insignificant disruption to the weights can destructively harm the quality of the models. Inspired by the ability of human brains to learn and consolidate knowledge through the waking-sleeping cycle, we propose a continual learning approach with two distinct phases: factorization and centralization, learning and merging knowledge accordingly. Our experiments on a sequence of varied code-switching datasets showed that the centralization stage can effectively prevent catastrophic forgetting by accumulating the knowledge in multiple scattering low-rank adapters.
arXiv.org Artificial Intelligence
Jun-23-2025
- Country:
- Asia > East Asia (0.04)
- Europe > Germany
- Baden-Württemberg > Karlsruhe Region > Karlsruhe (0.04)
- North America > United States
- Pennsylvania > Allegheny County > Pittsburgh (0.04)
- Genre:
- Research Report (1.00)
- Technology:
- Information Technology > Artificial Intelligence
- Machine Learning (1.00)
- Natural Language (1.00)
- Representation & Reasoning (1.00)
- Speech > Speech Recognition (1.00)
- Information Technology > Artificial Intelligence