Skip to Main content Skip to Navigation

Multi-level implementation of hierarchical neural models

Abstract : This paper addresses the following issue: Is event-based a relevant interface of engineering to model and provide solutions to the processing of biological signals? The first chapter deals with the setting up of an optical flow algorithm developed on event cameras inspired by the operation of the retina on the output signal of a rat retina. This work highlights the differences between the event model and the biology and exposes several adaptation solutions leading to an error in estimating the direction of flow of 3.5 ∞. The second chapter deals with the development of a real-time cell-sorting solution on the electrodes that record the output signal of a retina. The algorithm is evaluated by means of a data set known from the state of the art and manages to compete with non-real-time algorithms of the state of the art while more stable state make to the noise. Chapter 3 deals with the use of the algorithm presented chapter 2 in the project Sight Again to characterize ex-vivo the Prima prosthesis.The data treatments presented show the implant activation thresholds that are confirmed by the in-vivo part of the project. Chapter 4 deals with the generalization of the algorithm of Chapter 2 to the analysis of the variation of the pupillary contraction. This chapter highlights the biometric character of the pupil and uses hidden Markov models to study the pupil's behaviour. The document ends with a discussion about the limitations that event based applications may encounter, including the lack of compatible hardware platforms.
Document type :
Complete list of metadatas

Cited literature [99 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Thursday, June 11, 2020 - 4:32:45 PM
Last modification on : Thursday, August 20, 2020 - 4:29:07 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02865333, version 1


Kévin Géhère. Multi-level implementation of hierarchical neural models. Bioengineering. Sorbonne Université, 2018. English. ⟨NNT : 2018SORUS335⟩. ⟨tel-02865333⟩



Record views


Files downloads