Tesi etd-10242022-222557
Link copiato negli appunti
Tipo di tesi
Corso Ordinario Secondo Livello
Autore
BIASIZZO, MARCO
URN
etd-10242022-222557
Titolo
A recurrent deep neural network to simulate the early visual processing pathways
Struttura
Cl. Sc. Sperimentali - Ingegneria
Corso di studi
INGEGNERIA - INGEGNERIA
Commissione
relatore Prof. MICERA, SILVESTRO
Presidente Prof. AVIZZANO, CARLO ALBERTO
Membro Dott. LEONARDIS, DANIELE
Membro Prof. CIPRIANI, CHRISTIAN
Membro Prof. DI PASQUALE, FABRIZIO CESARE FILIPPO
Membro Prof. FORESTIERI, ENRICO
Membro Prof.ssa MENCIASSI, ARIANNA
Membro Prof. VITIELLO, NICOLA
Membro Prof. ABENI, LUCA
Membro Prof. BIONDI, ALESSANDRO
Membro Prof. CUCINOTTA, TOMMASO
Membro Dott.ssa COLLA, VALENTINA
Presidente Prof. AVIZZANO, CARLO ALBERTO
Membro Dott. LEONARDIS, DANIELE
Membro Prof. CIPRIANI, CHRISTIAN
Membro Prof. DI PASQUALE, FABRIZIO CESARE FILIPPO
Membro Prof. FORESTIERI, ENRICO
Membro Prof.ssa MENCIASSI, ARIANNA
Membro Prof. VITIELLO, NICOLA
Membro Prof. ABENI, LUCA
Membro Prof. BIONDI, ALESSANDRO
Membro Prof. CUCINOTTA, TOMMASO
Membro Dott.ssa COLLA, VALENTINA
Parole chiave
- CNN
- Conv-LSTM
- deep neural networks
- early visual pathway
- lateral geniculate nucleus
- spiking neural networks
- visual restoration
Data inizio appello
14/12/2022;
Disponibilità
completa
Riassunto analitico
We investigated the early visual pathways with a neuroscientific model representing the retina and the lateral geniculate nucleus.
We trained a recurrent, convolutional deep neural network to generate a firing rate activity consistent with the lateral geniculate nucleus behavior. The training dataset, composed by videos (inputs) and spiking times (targets), was generated through a physiological model we replicated from literature in pyNest. Then, we simulated an intraneural stimulation by imposing a custom activation to a selected central layer of the model.
Our results proved that recurrent, convolutional deep neural networks can be applied to simulate and analyze the early visual processing pathways.
We trained a recurrent, convolutional deep neural network to generate a firing rate activity consistent with the lateral geniculate nucleus behavior. The training dataset, composed by videos (inputs) and spiking times (targets), was generated through a physiological model we replicated from literature in pyNest. Then, we simulated an intraneural stimulation by imposing a custom activation to a selected central layer of the model.
Our results proved that recurrent, convolutional deep neural networks can be applied to simulate and analyze the early visual processing pathways.
File
| Nome file | Dimensione |
|---|---|
| Tesina_Santanna.pdf | 9.25 Mb |
Contatta l'autore |
|