An AlgorithmicBarrier to Neural Circuit Understanding

TL; DR: Non-existenceof tractable algorithms for neural circuit interrogation could pose a barrierto comprehensively understanding how neural circuits cause behavior.
Neuroscience is witnessing extraordinary progress in experimentaltechniques, especially at the neural circuit level. These advances are largelyaimed at enabling us to understand precisely how neural circuit computationsmechanistically cause behavior. Establishing this type of causal understandingwill require multiple perturbational (e.g optogenetic) experiments. It has beenunclear exactly how many such experiments are needed and how this number scaleswith the size of the nervous system in question. Here, using techniques fromTheoretical Computer Science, we prove that establishing the most extensivenotions of understanding need exponentially-many experiments in the number ofneurons, in many cases, unless a widely-posited hypothesis about computation isfalse (i.e. unless P = NP). Furthermore, using data and estimates, wedemonstrate that the feasible experimental regime is typically one where thenumber of experiments performable scales sub-linearly in the number of neuronsin the nervous system. This remarkable gulf between the worst-case and thefeasible suggests an algorithmic barrier to such an understanding. Determining whichnotions of understanding are algorithmically tractable to establish in whatcontexts, thus, becomes an important new direction for investigation.

Venkat Ramaswamy´s Worldwideneuro Theory Forum