Come & join the next #neuropizza!
Discover similarities between dreaming and methods applied in deep neural networks, join the interdisciplinary discussion, and eat the yummy pizza!

Many hypotheses of why brains dream have been proposed – recently the advent of deep neural networks provided the novel conceptual framework within this domain. Researches noted that both the brain and artificial neural networks face an omnipresent problem during learning: they tend to overfit to a particular dataset, which leads to an inability to generalize and perform properly on a novel, test dataset. Several analogies occur between how the brain faces the challenge of fitting too well to their daily distribution of stimuli and the methods and strategies applied in deep neural networks to optimize them. Evidence for the hypothesis is examined within neuroscience and deep learning and a set of testable predictions is introduced that can be tracked both in vivo and in silico.

The lecture will be given by Monika Pytlarz, a Sano student in the Brain and More Lab team.

Erik Hoel, The overfitted brain: Dreams evolved to assist generalization, Patterns, Volume 2, Issue 5, 2021, 100244, ISSN 2666-3899, https://lnkd.in/etsEzXMa. (https://lnkd.in/eN7zFpyk)