Sleep-Like Unsupervised Replay Improves Performance when Data are Limited or Unbalanced

arXiv:2402.10956v1 Announce Type: cross
Abstract: The performance of artificial neural networks (ANNs) degrades when training data are limited or imbalanced. In contrast, the human brain can learn quickly from just a few examples. Here, we investigated the role of sleep in improving the performance of ANNs trained with limited data on the MNIST and Fashion MNIST datasets. Sleep was implemented as an unsupervised phase with local Hebbian type learning rules. We found a significant boost in accuracy after the sleep …

abstract anns artificial artificial neural networks arxiv brain contrast cs.lg cs.ne data examples human learn networks neural networks performance role sleep training training data type unsupervised