David Oliver: Covid-19 and the Dunning-Kruger effect

1

Abstract

A cornerstone of theoretical neuroscience is the circuit model: a system of equations that captures a hypothesized neural mechanism. Such models are valuable when they give rise to an experimentally observed phenomenon – whether behavioral or in terms of neural activity – and thus can offer insights into neural computation. The operation of these circuits, like all models, critically depends on the choices of model parameters. Historically, the gold standard has been to analytically derive the relationship between model parameters and computational properties. However, this enterprise quickly becomes infeasible as biologically realistic constraints are included into the model increasing its complexity, often resulting in ad hoc approaches to understanding the relationship between model and computation. We bring recent machine learning techniques – the use of deep generative models for probabilistic inference – to bear on this problem, learning distributions of parameters that produce the specified properties of computation. Importantly, the techniques we introduce offer a principled means to understand the implications of model parameter choices on computational properties of interest. We motivate this methodology with a worked example analyzing sensitivity in the stomatogastric ganglion. We then use it to go beyond linear theory of neuron-type input-responsivity in a model of primary visual cortex, gain a mechanistic understanding of rapid task switching in superior colliculus models, and attribute error to connectivity properties in recurrent neural networks solving a simple mathematical task. More generally, this work suggests a departure from realism vs tractability considerations, towards the use of modern machine learning for sophisticated interrogation of biologically relevant models.