There was an intriguing exchange back and forth late 2015 that played out in the journal Nature Reviews Neuroscience which I only discovered early 2016. It started out with a review by Yuste which makes the case that neuroscience should expand beyond the “neuron doctrine”, placing the single neuron at the conceptual center of neuroscience, to a paradigm that includes neural networks more fully. The rationale is that networks “can form physiological units and generate emergent functional properties and states”.
Rubinov responded to this, pointing out that the general move towards network models is right, but raising the concern that the current state of neural network models “despite being enthusiastically researched at the end of the twentieth century, [..]] have largely not bridged the gap between elegant theory and neuroscientific observation”. To which I would strongly agree!
Yuste responded to this response with agreement: “There are many exciting areas of progress in current neuroscience detailing phenomenology that is consistent with some neural network models, some of which I tried to summarize and illustrate, but at the same time we are still far from a rigorous demonstration of any neural network model with causal experiments”. This is exactly what we are working on over at OpenWorm, so I am heartened to hear that there is interest in improving the state of neural networks in exactly this way.
This is part of my best of 2016 series to launch my new website. If you are interested, you can see the other posts.