Your browser does not support JavaScript!

Home    Collections    Type of Work    Post-graduate theses  

Post-graduate theses

Search command : Author="Παπαδάκης"  And Author="Αλέξανδρος"

Current Record: 5 of 5487

Back to Results Previous page
Next page
Add to Basket
[Add to Basket]
Identifier 000443669
Title Biologically constrained spiking neural network for image classification
Alternative Title Βιολογικά εμπνευσμένο spiking neural network για διαχωρισμό εικόνων
Author Μαλακάσης, Νικόλαος
Thesis advisor Ποϊράζη Παναγιώτα
Reviewer Τοπάλης, Παντελής
Παυλίδης, Παύλος
Abstract In the modern era of machine learning and artificial intelligence, artificial neural networks are widely used as the “brains” for smart software applications. Even though these algorithms are inspired by neurobiology, they lack features that neuronal circuits utilize to facilitate learning while employing learning rules that are not yet proven to be biologically plausible. In the presented study, we hypothesized that a biologically constrained neural network could perform image classification by utilizing network mechanisms and plasticity rules employed in the brain during learning. The network model includes an input layer and a single hidden layer that consists of multicompartmental spiking neurons with active dendrites, both excitatory pyramidals and inhibitory interneurons. Input neurons are sparsely connected to the hidden layer, and learning occurs by utilizing a class-specific learning signal and a synaptic tag-and-capture plasticity rule. We show that this biological network can perform binary image classification while maintaining the efficiency benefits of biological circuits. Moreover, this work highlights how structural plasticity improves performance by allowing the network to maximize its capacity. This model is arguably restricted due to computational inefficiency on regular hardware, which could be solved by implementing it on a neuromorphic platform. However, it can still serve as a template to test the advantages of biologically-inspired architectures when compared to deep learning algorithms, like its ability to learn continually. i
Language English
Subject Artificial intelligence
Brain
Dendrites
Learning
Δενδρίτες
Εγκέφαλος
Μάθηση
τεχνιτή νοημοσύνη
Issue date 2021-12-01
Collection   School/Department--School of Medicine--Department of Medicine--Post-graduate theses
  Type of Work--Post-graduate theses
Permanent Link https://elocus.lib.uoc.gr//dlib/8/6/4/metadata-dlib-1637572515-134066-30441.tkl Bookmark and Share
Views 37

Digital Documents
No preview available

No permission to view document.
It won't be available until: 2022-06-01