aiqbench analysis
This research demonstrates that Generative Meta-Continual Learning (GeMCL) enables stable few-shot spoken word classification across 1,000 classes with significantly reduced training time and data requirements compared to standard baselines. While the method does not consistently surpass fully-finetuned models in peak accuracy, its ability to adapt 2,000 times faster offers a compelling efficiency advantage for large-scale sequential learning tasks. The findings suggest that meta-learning approaches can effectively unlock the potential of larger-scale few-shot scenarios without sacrificing performance stability.
Scaling few-shot spoken word classification with generative meta-continual learning
arXiv:2605.13075v3 Announce Type: replace-cross Abstract: Few-shot spoken word classification has largely been developed for applications where a small number of classes is considered, and so the potential of larger-scale few-shot spoken word classification remains untapped. This paper investigates the potential of a spoken word classifier to sequentially learn to distinguish between 1000 classes when it is given only five shots per class. We demonstrate that this scaling capability exists by training a model using the Generative Meta-Continual Learning (GeMCL) algorithm and comparing it to repeatedly trained or finetuned baselines. We find that GeMCL produces exceptionally stable performance, and although it does not always outperform a repeatedly fully-finetuned HuBERT model nor a frozen HuBERT model with a repeatedly trained classifier head, it produces comparable performance to the latter while adapting 2000 times faster, having been trained less than half of the data for two orders of magnitude less time.
Abstract:Few-shot spoken word classification has largely been developed for applications where a small number of classes is considered, and so the potential of larger-scale few-shot spoken word classification remains untapped. This paper investigates the potential of a spoken word classifier to sequentially learn to distinguish between 1000 classes when it is given only five shots per class. We demonstrate that this scaling capability exists by training a model using the Generative Meta-Continual Learning (GeMCL) algorithm and comparing it to repeatedly trained or finetuned baselines. We find that GeMCL produces exceptionally stable performance, and although it does not always outperform a repeatedly fully-finetuned HuBERT model nor a frozen HuBERT model with a repeatedly trained classifier head, it produces comparable performance to the latter while adapting 2000 times faster, having been trained less than half of the data for two orders of magnitude less time.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2605.13075 [cs.CL]
(or arXiv:2605.13075v3 [cs.CL] for this version)
Full text under the publisher's open licence · 158 words
