Understanding self-motion induced optic flow processing in larval zebrafish in three levels

Abstract

In this thesis I present my PhD project about the neural implementation of self-motion induced optic flow processing in larval zebrafish pretectum, and dissect it with Marr's three-level framework for understanding visual system. I address the importance of this framework with a thought experiment and applied this framework to the optic flow processing problem. I state one of the computational goal(s) of self-motion induced optic flow processing is to extract the self-motion information from optic flow. The algorithms that can accomplish this goal, must implement two computational principles: the decomposability and identity. Then I briefly review three putative algorithms proposed for optic flow processing and discussed how the two computational principles are implemented in these algorithms. I also briefly review the relevant experimental studies and explained in detail that why the systematic RF characterization in flies may reveal the neural implementation of the matched filter algorithm for self-motion induced optic flow processing. Then I review the advantages and limitations of the reverse correlation method for RF estimation and I introduce the feature noise for efficiently estimating the RFs of advanced features.

I include three papers in the summary sections. The major results in my PhD project are in the first and the third paper while I mainly contribute to a specific technical problem in the second paper. In the discussion, I discuss the advantages and problems related to the spatiotemporal correlation of the CMN developed in the first paper. I discuss potential problems in the reverse correlation and the two-step NCB test used in the first and the third papers in terms of the effects in the interpretation of the RF estimation results. And I also discuss the limitation of reverse correlation technique in characterizing the functional properties whose linearity remains to be confirmed. Since the RFs in larval zebrafish do not cover the entire visual field which is deviated from the original “matched filter model”, I discuss how this deviation may affect the encoding quality and which computational principles are involved in. I also briefly mention the "mode-sensing" hypothesis and the pretectum-hindbrain problem in the downstream pathway in the sensorimotor transformation of optic flow. In the end, I discuss two specific gaps across the implementation, algorithmic and computational levels that my work failed to enclose. I hope this thesis may provide some insight into the optic flow processing and self-motion estimation in zebrafish, but also demonstrate how Marr's three-level framework may help to advance our understanding of visual system in a more systematic and rigorous manner.