Can Linear-Nonlinear Models capture simple forms of retinal adaptation? - Inria - Institut national de recherche en sciences et technologies du numérique
Poster Année : 2024

Can Linear-Nonlinear Models capture simple forms of retinal adaptation?

Résumé

The retina is a tissue situated in the inner part of the eyes of vertebrates. It is responsible for transforming the light of the visual field into action potentials, forwarded to the visual cortex through the optic nerve. Short-term Plasticity (STP), the strengthening or weakening of synapses in reaction to the past pre- and post-synaptic activity, plays an important role in this network's ability to encode an ever-changing visual environment, for example across different levels of luminosity and contrast. While the retina's activity in response to static or uncorrelated inputs is commonly described through Linear-Nonlinear (LN) Models, these models often fail to capture responses to dynamic stimuli with spatiotemporal correlations such as natural movies. In this project, we apply different versions of the classical LN model to Retinal Ganglion Cells – the output stream of the retina – and test its performance on their responses to a simple stimulus protocol designed to measure STP: a sequence of repeated light flashes. We show that these models fail as well to reproduce STP as captured in experiments. The Linear-Nonlinear Model comprises two components. Firstly, a stimulus is passed through a linear filter, representing the neuron's temporal receptive field. Secondly, the output is transformed by a nonlinear function to predict the instantaneous firing rate of each ganglion cell. For both components, either a nonparametric (raw) function can be computed, or a smooth, parametric curve fitted to the former may be used. The smooth kernels investigated here include a biphasic difference of exponentials and a difference of Gaussians. We obtain raw estimates for the filter and the nonlinearity from responses to white-noise stimulation and optimise the smooth LN-model for cells that exhibit forms of short term plasticity, ie. those in which repetitions of the same flash lead to a gradual increase or decrease in response firing rate. We find that these neurons are best fitted by a bi- or triphasic linear filter (ie. with a local maximum followed by a minimum) and a sigmoidal nonlinearity. Although the LN-model correctly predicts the timing of retinal responses to the stimulus, in this particular case it does not capture the forms of adaptation tested, eg. an increase in response between the first and second flash of the stimulus. In order to describe STP, further investigation is needed including STP implementation in a model of the retinal network.
Fichier principal
Vignette du fichier
neuromod_poster_piovano.pdf (1.07 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04708984 , version 1 (25-09-2024)

Licence

Identifiants

  • HAL Id : hal-04708984 , version 1

Citer

Laura Piovano, Simone Ebert, Bruno Cessac. Can Linear-Nonlinear Models capture simple forms of retinal adaptation?. NeuroMod 2024 - Annual Meeting of the Neuromod Institute, Jul 2024, Antibes, France. ⟨hal-04708984⟩
0 Consultations
0 Téléchargements

Partager

More