Simulating prescribed error sensitivi

Nikonov, Arseniy (2020) Simulating prescribed error sensitivi. Bachelor's Thesis, Artificial Intelligence.

Preview	Text AI_BA_2020_ARSENIY_NIKONOV.pdf Download (623kB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (98kB)

Abstract

: Memristors are devices that are part of an electrical circuit and that act as resistors and can control the flow of current through the system. By controlling the polarity and magnitude of a voltage we can set the device into multiple resistance states. It was theorized that these devices can be used for neuromorphic computations. Artificial neural networks are collection of connected nodes. This connections represent synapses of biological brains. Memristor properties can be used to emulate synapses. In this research, the simulation of prescribed error sensitivity(PES) learning was built using memristive devices as connectors between the neurons. Using memristors two different networks were built and compared to the PES network without the use of memristors. In the first network, memristors were initialized in the highest resistance state while in the second the initial resistances were randomized between certain values. Both networks were able to learn the identity function but were slower than the non-memristive simulation. For the second experiment, we investigated which values of the initial resistance the simulation can work successfully and reached the conclusion that the minimal starting resistance should be above 106 Ohm. From this experiment, we conclude that memristors can be used effectively to emulate synapses.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Borst, J.P. and Taatgen, N.A.
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	31 Jan 2020 14:47
Last Modified:	31 Jan 2020 14:47
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/21488

Actions (login required)

View Item