Izhikevich neural model and STDP learning algorithm mapping on spiking neural network hardware emulator
Visualitza/Obre
Estadístiques de LA Referencia / Recolecta
Inclou dades d'ús des de 2022
Cita com:
hdl:2117/350516
Tipus de documentProjecte Final de Màster Oficial
Data2020-11-15
Condicions d'accésAccés obert
Tots els drets reservats. Aquesta obra està protegida pels drets de propietat intel·lectual i
industrial corresponents. Sense perjudici de les exempcions legals existents, queda prohibida la seva
reproducció, distribució, comunicació pública o transformació sense l'autorització del titular dels drets
Abstract
From the 20th century, biological mechanisms of the brain behaviour have become more and more interesting for the research communities in information fields due to the computational power of the systems they inspire. In fact, despite the lack of consensus about the information processing actually involved in brain, biological processes have served as reference for recent computational models. The first Artificial Neural Networks (ANNs) were developed as simplified versions of biological neural net- works in terms of structure and function. Today, the third generation of artificial network is that of the Spiking Neural Networks (SNNs), which reach a more realistic modelling by utilizing true biological features, like spikes, to transmit information between neurons. The proposal of this thesis is to embed the Izhikevich neuron model and a full custom "Spike timing dependent plasticity" (STDP) learning algorithm in an architecture called HEENS (Hardware Emulator of Evolved Neural System). HEENS is a multi-chip structure developed at the "Universitat Politecnica de Catalunya" (UPC) based on a ring link topology connecting several SIMD processors reproducing each one a group of neuron of a Spiking neural network. The Izhikevich neuron model is a worldwide adopted mathematical model for reproducing the neural membrane potential evolution, observed in some mammalian cortex, a long time and according to external stimuli. STDP is a biological learning algorithm which shapes the strength of a synaptic connection according to the timing with which that connection takes part to the overall spiking activity of the post or pre-synaptic neurons. This master thesis project, in particular, acts at algorithm level and at instruction level as well at architectural level. It takes place analysing the mathematical models for the right data parallelism, writing the assembly program describing the routine common to all the neurons of the implemented SNN, modifying the instruction set and the existing hardware of the HEENS architecture, in order to fullfil the biological model needs from a computational and performance point of view. HEENS architecture is described in VHDL code, its set-up operations (assembler for code translation, generation of memories, Network configuration) are performed by Python scripts, the comparison between the actual behaviour of HEENS to that of the mathematical models is instead performed via MatLAB scripts. The latter allow: to imitate the performances of a special purpose hardware; to generate source files in order to synchronize and align the model and the architecture even with the randomization of several neural parameters; to make some design choices; to verify and to show the results.
Col·leccions
Fitxers | Descripció | Mida | Format | Visualitza |
---|---|---|---|---|
Caruso_Master_thesis.pdf | 3,874Mb | Visualitza/Obre |