Ubichip Virtual Machine and visualization of Spiking Neural Network Parameters
Tutor / Supervisor
Student
Budzisz, Jan Maciej
Document type
Master thesis (pre-Bologna period)
Date
2010
rights
Open Access
Publisher
Universitat Politècnica de Catalunya
UPCommons
Abstract
Since ages people have been wondering the inner workings of brain.
There is more and more information about processes involving particular
brain regions, yet synthesizing neural-networks in a large scale has always
been a problem. Despite the fact that the popular desktop computer
is becoming more powerful, simulating massively parallel computations
aways proved to be a challenge. Due to its genericness, in very specialized
work, such as neural networks, it is far better to design and use dedicated
array of processors.
Such approach has been used in development of the Ubichip – a device
designed specifically for such purposes. The design goal was to simplify the
execution elements to suit the needs of efficient neuron model algorithms
emulation.
As with every development cycle of a complex tool there are many tasks
that may be carried out in parallel. This, of course, effects in allowing
the development team to shorten the cycle and provide the solution faster.
This however, requires a set of tools that will allow to prototype and verify
work progress against some set of basic rules.
The Ubichip was already supported by a toolkit named SpiNDeK,
that allowed to create networks and with the use of ModelSim simulate
the code. This solution for proof-checking the algorithms however proved
to be cumbersome and due to many levels of indirection – slow. To increase
efficiency when working with code, the programmer should not
have to wait endlessly watching the progress bars. To remedy this, improve
efficiency and encourage more precise tuning and development of
new neuron-model algorithms the Ubichip Virtual Machine was born.
At first it was just meant to be a simple visualization tool, but as it
turned out there was a missing link in the chain of tools available for the
Ubichip, which had to be filled.
Thus, in this work a virtual machine for the Ubichip has been developed,
as well as a visualization tool that enables a convenient display of
the evolution of spiking neurons in a network.
i
