Graphical representation of data for a multiprocessor platform emulating spiking neural networks
Tutor / director / evaluatorMadrenas Boadas, Jordi
Document typeMaster thesis (pre-Bologna period)
Rights accessOpen Access
Research in the eld of simulating large-scale spiking neural networks (SNN) has been carried out within the frame of Perplexus a European-funded re- search project based on a university consortium. In this project, a semi- custom electronic device called Ubichip has been designed. The mode of interest of this chip to emulate SNNs is based on a SIMD (Single-Instruction Multiple-Data) multiprocessor machine. The software for generating the as- sembly containing simulation of Iglesias-Villa spiking neural network model was also developed within that project and it is currently being successfully used for running neural network emulation on Ubichip. The tools developed so far are useful for debugging by simulation, but in order to evaluate the behavior of SNN being emulated, two needs arose: real- time monitoring of the network evolution and a higher-level, understandable visualization solution. First, the existing software that was developed in the Perplexus project has been analyzed. After examining all available solutions, including writing a standalone dedicated program, it was nally decided to develop the so-called Ubiplot plug-in. The reason was to take advantage of the existing Ubimanager environment. The development started by verifying the communication with the Ubichip, so simple waveforms for data in a given address in the Ubichip's RAM were implemented. Then the plug-in was extended with histogram and raster plots that are accessing multiple locations of the memory in each execution step. This led to the creation of the variable map that de nes the program's variables and their precise placement in the RAM. At the end simple logging facility and possibility to save and restore the layout of the plots were added. This thesis describes the Ubiplot and the development e ort put in its creation.