Thesis Project Form
Title (tentative): A neuromorphic system for learning stereopsis in active settings| Thesis advisor(s): Sabatini Silvio P., Andrea Canessa, Francesca Peveri | E-mail: |
| Address: Via All'Opera Pia, 13 - 16145 Genova (III piano) | Phone: (+39) 010 33 52092 |
Description
Motivation and application domain
Stereopsis requires precise vergence control and balanced retinocortical convergence of visual signals coming from the two eyes. Absence of sensorimotor disorders during active vision experience in a critical period is crucial for a correct development.
General objectives and main activities
The thesis aims at training a well-established model of early representation of information in the human visual cortex, to (1) investigate the necessary and sufficient conditions that lead to typical development, (2) advance hypotheses on conditions leading to atypical development, (3) predict the effects of corrective counteractions.
The starting point is (1) a “pre-trained” front-end “convolutional” neural network (working with monocular spatiotemporal receptive fields on the signal provided by the two retinas), (2) a cortical circuitry that governs redundant binocular convergence, and (3) a kinematic model of the binocular oculomotor system. Learning should not require any supervision from outside but rely on some form of self-calibration.
The starting point is (1) a “pre-trained” front-end “convolutional” neural network (working with monocular spatiotemporal receptive fields on the signal provided by the two retinas), (2) a cortical circuitry that governs redundant binocular convergence, and (3) a kinematic model of the binocular oculomotor system. Learning should not require any supervision from outside but rely on some form of self-calibration.
Training Objectives (technical/analytical tools, experimental methodologies)
The student will learn to employ an array of methodologies and instrumentation, including:
• Stereoscopic rendering and binocular image formation
• Firing rate neuromorphic models
• Python/MATLAB programming language
• Unsupervised learning
• Stereoscopic rendering and binocular image formation
• Firing rate neuromorphic models
• Python/MATLAB programming language
• Unsupervised learning
Place(s) where the thesis work will be carried out: Bioiengineering Lab (PSPC), via Opera Pia 13
Additional information
Pre-requisite abilities/skills: Neuromorphic computing
Maximum number of students: 2
Financial support/scholarship: None