Evolvable Block-Based Neural Network Design for Applications in Dynamic Environments

Abstract

Dedicated hardware implementations of artificial neural networks promise to provide faster, lower-power operation when compared to software implementations executing on microprocessors, but rarely do these implementations have the flexibility to adapt and train online under dynamic conditions. A typical design process for artificial neural networks involves offline training using software simulations and synthesis and hardware implementation of the obtained network offline. This paper presents a design of block-based neural networks (BbNNs) on FPGAs capable of dynamic adaptation and online training. Specifically the network structure and the internal parameters, the two pieces of the multiparametric evolution of the BbNNs, can be adapted intrinsically, in-field under the control of the training algorithm. This ability enables deployment of the platform in dynamic environments, thereby significantly expanding the range of target applications, deployment lifetimes, and system reliability. The potential and functionality of the platform are demonstrated using several case studies.

Evolvable Block-based Neural Networks for classification of driver drowsiness based on heart rate variability

“…In most cases, this additional feature is not needed [5], and can be safely set downwards in direction. This will cause the BbNN to fall into the feed-forward neural network category.…”

“…Based on previous works, GA is the most popular and well documented tool for this purpose [5]. The complete GA methodology is well described in previous works, such as in [17].…”

“…Block-based neural networks (BbNN) provides a more unified solution to the two fundamental problems of ANN, which are simultaneous optimization of structure, and implementation using digital hardware such as FPGAs. BbNNs are trained using genetic algorithm (GA), and have been successfully deployed in the past for various kinds of classification problems [5].…”

“…LITERATURE REVIEW AND RELATED WORKS Ever since ANNs were introduced, it has always been criticized for inefficiency [4], [5], [6]. To implement large and effective neural networks, much processing and storage resources need to be committed.…”

“…Training is done with the help of an evolutionary algorithm (EA), such as genetic algorithm (GA). BbNNs have been successfully used in various applications such as ECG signal classification [8], hypoglycemia detection [9], pattern recognition [10], and many more [5].…”