A Computational Study of the Role of Neurogenesis in Learning and Memory

Neurogenesis persists throughout life in the dentate gyrus of the mammalian hippocampus. There is a large body of evidence that neurogenesis may impact cognition, particularly concerning learning and memory, though it is unclear what mechanisms underlie these processes. Computational models have established that the addition of neurons degrades existing memories (i.e., produces forgetting). These predictions are supported by empirical observations in rodents, where post-training increases in neurogenesis also promote forgetting of hippocampus-dependent memories. However, in these models that use 10-1,000 neurons to represent the dentate gyrus, forgetting is only observed at new neuron addition rates that greatly exceed adult neurogenesis rates observed in vivo. To address this, we generated an artificial neural network that incorporated more realistic features of the hippocampus – including increased network size, sparse activity, and sparse connectivity and explored how these properties modulate the impact of using biologically relevant rates of neurogenesis in a pattern categorization task. Our model captures several mnemonic phenomena associated with neurogenesis, including forgetting and enhanced reversal learning. These effects were sensitive to changes in increased output connectivity and excitability of new neurons. Crucially, forgetting was observed at lower rates of neurogenesis in larger networks, with the addition of as little as 0.1% of the total DG population sufficient to induce forgetting. We also propose a role for neurogenesis in the generalization of memories using traditional deep learning architectures. At rates of 3% turnover, we found that neurogenesis alone improved generalization on unseen data in two commonly used categorization tasks, the MNIST and CIFAR-10 datasets. Together, we suggest not only that low rates of neurogenesis can have an impact on cognition, but that we can use biologically-inspired and artificial neural networks to investigate the computational functions of neurogenesis that may mirror those occurring in vivo as well.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.