Spiking Neural Networks (SNNs), also known as the third generation of neural networks, are considered a more biologically plausible approach of artificial neural networks as they model the information transfer and processing as occurs in biological neurons, i.e. via the precise timing of spikes (discrete events at points in time). One of the main advantages of this type of neural networks is that they can exploit the benefits of neuromorphic implementation, such as ultra-low-power consumption and real-time processing. These entail advantages such as the greater autonomy of the devices on multiple application fields such as robotics, automotive and wearable and portable devices for sport and health monitoring, together with the environmental benefits due to negligible power consumption.
Despite all these advantages, SNN field is not yet mature enough. Among the improvement areas, temporal information and time series processing can be found. However, little effort has been made to solve regression problems by SNN such as Time-Evolution Modelling, which are very common in industrial and bioengineering applications. Actually, SNN have been mostly used to date for troubleshooting classification. One of the main reasons for this is poor decoding quality provided by the state-of-the-art encoding algorithms.

In this respect, ViSens has proposed the design and development of a new encoding and decoding algorithm inspired in the well-known pulsewidth modulation (PWM) which enables to encode analog signals into spike sequences, and decode spikes with very high/user-required reconstruction accuracy through one simple parameter. Thus, the aim of this research project is to take a step forward in regression problems solving by proper training  of SNN in conjunction with PWM based encoding/decoding algorithm.

A HardWare Implementation of the PWM inspired encoding/decoding algorithm (IHA-PWM).
A new supervised training approach of SNN for regression based on both backpropagation and PWM inspired encoding/decoding algorithm.
Define the specification for a neuromorphic HardWare implementation of both IHA-PWM and the new supervised training algorithm.

Amongst the achievements of the project, we can highlight the following:

• A first HardWare prototipe based on Arduino UNO.
• Design and development an extension for compression based on the PWM inspired encoding/decoding algorithm.
• Progress on supervised training of benchmark signals.