Bioloogiliselt realistliku neuroni mudeli ehitamine ja uurimine
Date
2015
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
On näidatud, et kaltsiumiioonide tööl põhinevad üksikneuroni mudelid võimaldavad erinevate sünaptilise plastilisuse vormide teket. Käesolevas töös koostati üks selline mudel ja uuriti mudeli õppimisvõimet.
Koostatud mudeli käitumine kordas kvalitatiivselt varasemate uurimuste tulemusi, välja arvatud korreleeritud sisendmustrite korral. Leiti, et neuron töötab lineaarse filtrina, kuna õpitud sisendmustrit vaid osaliselt nähes sõltub väljund lineaarselt õpitud mustri osakaalust. Tõenäosuslik virgatsaine vabanemine vähendas oodatult sisendmustrite korrelatsiooni ja parandas infoülekande tõhusust. Peakomponentanalüüsiks neuron töös formuleeritud viisil võimeline ei olnud. Tulemuste sõltuvust parameetrite täpsetest väärtustest ei kontrollitud.
Neuroni võime keerulisemat infotöötlust teha ei leidnud tõestust. Hoolimata sellest on koostatud neuroni mudel võimeline kasulikuks infotöötluseks ja seega hea alus edasisteks uuringuteks.
Calcium-based single neuron models have been shown to elicit different modes of synaptic plasticity. In the present study one such model was implemented and its learning behaviour studied. Behaviour of the implemented neuron agreed qualitatively with prior work in all regards except selectivity to correlation in input. The neuron was found to implement a linear filter responding linearly to partial presentations of learned patterns. Simulating probabilistic neurotransmitter release had an expected effect of de-correlating input and was found to improve the efficiency of information transfer. In the regimes explored, the neuron was found to be incapable of performing principal component analysis. The insensitivity of results to changes in parameters was mostly untested. The neuron did not exhibit more advanced information processing capabilities in the tests conducted. However, the implemented neuron model is capable of meaningful information processing and forms a good basis for further research.
Calcium-based single neuron models have been shown to elicit different modes of synaptic plasticity. In the present study one such model was implemented and its learning behaviour studied. Behaviour of the implemented neuron agreed qualitatively with prior work in all regards except selectivity to correlation in input. The neuron was found to implement a linear filter responding linearly to partial presentations of learned patterns. Simulating probabilistic neurotransmitter release had an expected effect of de-correlating input and was found to improve the efficiency of information transfer. In the regimes explored, the neuron was found to be incapable of performing principal component analysis. The insensitivity of results to changes in parameters was mostly untested. The neuron did not exhibit more advanced information processing capabilities in the tests conducted. However, the implemented neuron model is capable of meaningful information processing and forms a good basis for further research.