Abstract

The majority of cortical neurons have an exclusive physiological effect, each neuron either excites or inhibits all its synaptic targets. This empirical observation, known as Dale’s principle, poses a strong constraint on the structure and computation of neural circuits, but its function remains unclear. Proposed explanations include: balancing of neural activity, generating oscillations, accelerating and/or amplifying responses to stimuli; However, Dale’s principle is not necessary for implementing any of these functions. A seemingly unrelated, but influential theory, posits that neural circuits aim at decomposing input stimuli into a set of uncorrelated variables, called latent factors. Here I propose that neural circuits aim at transmitting, rather than reconstructing, those latent factors. I show that Dale’s principle is necessary and sufficient for transmitting latent factors, and that known instances of synaptic plasticity rules (STDP) are consistent with its implementation. I study a dynamical model of a neural circuit characterized by a given synaptic matrix and input covariance matrix. I show that the latent factors of the input covariance can be transmitted by the neural circuit provided that the synaptic matrix satisfies a set of constraints. I show that for most types of inputs, those constraints tend to enforce Dale’s principle. Conversely, I show that enforcing Dale’s principle in the synaptic matrix tends to satisfy those constraints. I construct a synaptic plasticity rule that implements these constraints and I show that is consistent with STDP . These results suggest a novel organizational principle of neural circuits that will help predicting their behavior. Future work will be devoted to studying how the latent factors are transmitted in order to solve specific tasks.