What is the state of neuroscience?

Most recurrent neural network models of the brain assume that the strength of neuronal connections follows a normal distribution, a pattern known as Gaussian connectivity. But a new computational simulation adds to mounting evidence that these connections instead show a heavy-tailed distribution, in which a few connections are particularly strong, and others are weaker. Heavy-tailed synaptic weight distributions, unlike Gaussian ones, allow neural networks to balance flexibility and stability, the study finds. That’s because they are more likely to stay at the “edge of chaos,” where they can adapt without becoming unstable at a wide range of parameters, says study investigator Eva Yi Xie, a visiting scientist at the Allen Institute. The research “helps to further establish the need to think beyond Gaussian distributions for synaptic weights,” says Blake Richards, associate professor of computational neuroscience at McGill University, who was not involved in the study. — Natalia Mesa

Aidan J. Horner argues that existing systems consolidation theories do not effectively capture the dynamism of episodic memory, and he proposes a “3D state space for episodic memories,” in which two dimensions relate to retrieval driven by the hippocampus or the neocortex, and another dimension relates to the specificity of the memory.