Introduction
When climate change becomes more and more prevalent around the world, reducing power consumption of electronic devices, as a result, gets the close attention of the technological research. One of the research fields suggests the neuromorphic processor as one of the possible solutions. The processor has the potential to overcome the power bottleneck of the traditional Von Neumann architecture, by emulating the way the biological brain processes information every day. It is noticeable that the biological system consumes significantly low power to accomplish complex cognitive tasks, therefore, the brain-like (i.e. neuromorphic) processor is predicted to perform similarly in terms of power consumption nature. For example, bees can process many complex cognitive functions and dissipate only around 10uW, while the DARPA robot managing similar functionality consumes roughly 1kW [1]. The key to the power-saving feature is that the biological signals are generated and propagated as action potentials (i.e and impulse/ spike signal) and driven by event-based mechanism [1]. On the other hand, there have been vast amount of Artificial Intelligence (AI) applications, such as Perceptron network, Artificial Neural Network with backpropagation [2], Convolution Neural Network, etc. However, these approaches do not follow the spike-based process. Therefore, it is intriguing to compare them with a spike-based network, so-called Spiking Neural Network (SNN).
Project description
The objective of this project is getting accustomed to neuromorphic field, simulate a Spiking Neuro Network (SNN) and compare its performance with an Artificial Neuro Network (ANN). The proposed design / system / device of the project has the following goals:
Goal: To develop an SNN and a comparable other AI application for a specific task.
Required skills
Experience with Python, computer algorithms is preferred.
References
[1] Liu, S., & Delbruck, T. (2010). Neuromorphic sensory systems. Current Opinion in Neurobiology, 20(3), 288–295. https://doi.org/10.1016/j.conb.2010.03.007
[2] Rumelhart, David E. et al. “Learning representations by back-propagating errors.” Nature 323 (1986): 533-536.
