Spiking neural Networks. On-Line learning in Event based Neuromorphic Systems

• Autores: Ajay Vasudevan
• Directores de la Tesis: Bernabe Linares Barranco (dir. tes.), María Teresa Serrano Gotarredona (dir. tes.)
• Lectura: En la Universidad de Sevilla ( España ) en 2025
• Idioma: inglés
• Número de páginas: 145
• Enlaces
  • Tesis en acceso abierto en: Idus
• Resumen

  • Spiking Neural Networks are considered as the third generation of neural networks. As opposed to conventional Artificial Neural Networks where temporal information is not explicitly considered, Spiking Neural Networks incorporate time as an explicit variable which enriches the spatio-temporal representation of the signal. In Spiking Neural Networks, signals are represented and communicated as a flow of spikes. Consequently, communication and computation power is required only during the occurrence of a spike. With proper sparse coding of the signals, SNNs offer advantages over other neural networks in terms of processing speed and power requirements. However, while training techniques for conventional neural networks which make use of techniques like Gradient Descent and Back Propagation are very mature and achieve high levels of accuracy in supervised classification problems, the training of SNNs in a supervised manner is still much less developed. Although the explicit incorporation of temporal information has more potential to solve problems where dynamics plays an important role, it also introduces complexity in the training. Training in an unsupervised manner can be done with local learning rules which are biologically plausible but these unsupervised techniques are still achieving less accurate results. This thesis details the background and motivations for using SNNs and presents work done of training SNNs in both unsupervised and supervised manners.