2018 Conference on Cognitive Computational Neuroscience

5-8 September 2018    Philadelphia, Pennsylvania

Technical Program

Paper Detail

Paper: PS-1B.2
Session: Poster Session 1B
Location: Symphony/Overture
Session Time: Thursday, September 6, 18:45 - 20:45
Presentation Time:Thursday, September 6, 18:45 - 20:45
Presentation: Poster
Publication: 2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania
Paper Title: A Procedural Roadblock to Mechanistic Understanding of Neural Circuits
Manuscript:  Click here to view manuscript
DOI: https://doi.org/10.32470/CCN.2018.1159-0
Authors: Venkat Ramaswamy, National Centre for Biological Sciences, India
Abstract: Neuroscience is witnessing impressive progress in techniques for observing and interrogating neural circuits. Advances include optical readout of neural circuit activity, capability to optogenetically stimulate/silence subsets of neurons in-vivo and ascertaining exact anatomical connectivity, for increasingly larger neural circuits. It is thought that progress in such technologies holds promise in ultimately enabling us to understand mechanistic computation in neural circuits leading to behavior. Here, using techniques from Theoretical Computer Science, we examine how many experiments are needed to establish an empirical understanding of mechanistic circuit computation, for a fixed behavior. It is proved, mathematically, that establishing the most extensive notions of understanding {\em needs} exponentially-many experiments in the number of neurons, in general, unless a widely-posited hypothesis about computation is false. To make matters worse, the feasible experimental regime is one where the number of experiments scales sub-linearly in the number of neurons. Together, this suggests that such a comprehensive understanding is de facto unknowable, in general. Determining which notions of understanding are algorithmically tractable, thus, becomes an important direction for investigation. A similar roadblock may exist in our quest to comprehensively understand contemporary deep neural networks as well.