Weightless Neural Models: A Review of Current and Past Works

This paper presents a survey of a class of neural models known as Weightless Neural Networks (WNNs). As the name suggests, these models do not use weighted connections between nodes. Instead, a dierent kind of neuron model, usually based on RAM memory devices, is used. In the literature, the terms \RAM-based " and \n-tuple based " systems are also commonly used to refer to WNNs. WNNs are being widely investigated, motivating relevant applications and two international workshops in the last few years. The paper describes the most important works in WNNs found in the literature, pointing out the challenges and future directions in the area. A comparative study between weightless and weighted models is also presented.