A Dynamically Reconfigurable BbNN Architecture for Scalable Neuroevolution in Hardware

In this paper, a novel hardware architecture for neuroevolution is presented, aiming to enable the continuous adaptation of systems working in dynamic environments, by including the training stage intrinsically in the computing edge. It is based on the block-based neural network model, integrated with an evolutionary algorithm that optimizes the weights and the topology of the network simultaneously. Differently to the state-of-the-art, the proposed implementation makes use of advanced dynamic and partial reconfiguration features to reconfigure the network during evolution and, if required, to adapt its size dynamically. This way, the number of logic resources occupied by the network can be adapted by the evolutionary algorithm to the complexity of the problem, the expected quality of the results, or other performance indicators. The proposed architecture, implemented in a Xilinx Zynq-7020 System-on-a-Chip (SoC) FPGA device, reduces the usage of DSPs and BRAMS while introducing a novel synchronization scheme that controls the latency of the circuit. The proposed neuroevolvable architecture has been integrated with the OpenAI toolkit to show how it can efficiently be applied to control problems, with a variable complexity and dynamic behavior. The versatility of the solution is assessed by also targeting classification problems.

I-Vector-Based Patient Adaptation of Deep Neural Networks for Automatic Heartbeat Classification

Automatic classification of electrocardiogram (ECG) signals is important for diagnosing heart arrhythmias. A big challenge in automatic ECG classification is the variation in the waveforms and characteristics of ECG signals among different patients. To address this issue, this paper proposes adapting a patient-independent deep neural network (DNN) using the information in the patient-dependent identity vectors (i-vectors). The adapted networks, namely i-vector adapted patient-specific DNNs (iAP-DNNs), are tuned toward the ECG characteristics of individual patients. For each patient, his/her ECG waveforms are compressed into an i-vector using a factor analysis model. Then, this i-vector is injected into the middle hidden layer of the patient-independent DNN. Stochastic gradient descent is then applied to fine-tune the whole network to form a patient-specific classifier. As a result, the adaptation makes use of not only the raw ECG waveforms from the specific patient but also the compact representation of his/her ECG characteristics through the i-vector. Analysis on the hidden-layer activations shows that by leveraging the information in the i-vectors, the iAP-DNNs are more capable of discriminating normal heartbeats against arrhythmic heartbeats than the networks that use the patient-specific ECG only for the adaptation. Experimental results based on the MIT-BIH database suggest that the iAP-DNNs perform better than existing patient-specific classifiers in terms of various performance measures. In particular, the sensitivity and specificity of the existing methods are all under the receiver operating characteristic curves of the iAP-DNNs.

Evolvable rough-block-based neural network and its biomedical application to hypoglycemia detection system

This paper focuses on the hybridization technology using rough sets concepts and neural computing for decision and classification purposes. Based on the rough set properties, the lower region and boundary region are defined to partition the input signal to a consistent (predictable) part and an inconsistent (random) part. In this way, the neural network is designed to deal only with the boundary region, which mainly consists of an inconsistent part of applied input signal causing inaccurate modeling of the data set. Owing to different characteristics of neural network (NN) applications, the same structure of conventional NN might not give the optimal solution. Based on the knowledge of application in this paper, a block-based neural network (BBNN) is selected as a suitable classifier due to its ability to evolve internal structures and adaptability in dynamic environments. This architecture will systematically incorporate the characteristics of application to the structure of hybrid rough-block-based neural network (R-BBNN). A global training algorithm, hybrid particle swarm optimization with wavelet mutation is introduced for parameter optimization of proposed R-BBNN. The performance of the proposed R-BBNN algorithm was evaluated by an application to the field of medical diagnosis using real hypoglycemia episodes in patients with Type 1 diabetes mellitus. The performance of the proposed hybrid system has been compared with some of the existing neural networks. The comparison results indicated that the proposed method has improved classification performance and results in early convergence of the network.

Hierarchical Rank Density Genetic Algorithm for Radial-Basis Function Neural Network Design

In this paper, we propose a genetic algorithm based design procedure for a radial-basis function neural network. A Hierarchical Rank Density Genetic Algorithm (HRDGA) is used to evolve the neural network's topology and parameters simultaneously. Compared with traditional genetic algorithm based designs for neural networks, the hierarchical approach addresses several deficiencies highlighted in literature. In addition, the rank-density based fitness assignment technique is used to optimize the performance and topology of the evolved neural network to deal with the confliction between the training performance and network complexity. Instead of producing a single optimal solution, HRDGA provides a set of near-optimal neural networks to the designers so that they can have more flexibility for the final decision-making based on certain preferences. In terms of searching for a near-complete set of candidate networks with high performances, the networks designed by the proposed algorithm prove to be competitive, or even superior, to three other traditional radial-basis function networks for predicting Mackey–Glass chaotic time series.

Block based neural network for hypoglycemia detection

In this paper, evolvable block based neural network (BBNN) is presented for detection of hypoglycemia episodes. The structure of BBNN consists of a two-dimensional (2D) array of fundamental blocks with four variable input-output nodes and weight connections. Depending on the structure settings, each block can have one of four different internal configurations. To provide early detection of hypoglycemia episodes, the physiological parameters such as heart rate (HR) and corrected QT interval (QTc) of electrocardiogram (ECG) signal are used as the inputs of BBNN. The overall structure and weights of BBNN are optimized by an evolutionary algorithm called hybrid particle swarm optimization with wavelet mutation (HPSOWM). The optimized structures and weights of BBNN are capable to compensate large variations of ECG patterns caused by individual and temporal difference since a fixed structure classifiers are easy to fail to trace ECG signals with large variations. The ECG data of 15 patients are organized into a training set, a testing set and a validation set, each of which has randomly selected 5 patients. The simulation results shows that the proposed algorithm, BBNN with HPSOWM can successfully detect the hypoglycemic episodes in T1DM in term of testing sensitivity (76.74%) and test specificity (50.91%).

Atmospheric degradation correction of terahertz beams using multiscale signal restoration

We present atmospheric degradation correction of terahertz (THz) beams based on multiscale signal decomposition and a combination of a Wiener deconvolution filter and artificial neural networks. THz beams suffer from strong attenuation by water molecules in the air. The proposed signal restoration approach finds the filter coefficients from a pair of reference signals previously measured from low-humidity conditions and current background air signals. Experimental results with two material samples of different chemical compositions demonstrate that the multiscale signal restoration technique is effective in correcting atmospheric degradation compared to individual and non-multiscale approaches.

Implementation of a Real-Time FPGA-Based Intelligent Parallel Parking System

Parallel parking is a challenging maneuver for many drivers, especially in crowded cities with heavy traffic congestion and limited parking spaces. This research focuses on the development of an intelligent parallel parking system. Conventional vehicular control techniques generally require the use of analytical models. However, complexities due to nonlinear car dynamics create multiple challenges when modeling car motion. As fuzzy logic control is appropriate for nonlinear and complex systems where human expert knowledge is available, it is well-suited for the parking application. This paper presents the design and implementation of a real-time fuzzy logic based parallel parking system. Two control units consisting of a main controller and a secondary fuzzy logic controller are utilized. The latter controller is employed for the realization of the wall following task, which has an important role in the parking system. Based on performance and flexibility considerations, the control units are implemented onto a reconfigurable hardware platform, namely a Field Programmable Gate Array (FPGA). A prototype vehicle is developed to ensure the proposed algorithm provides vehicular systems with the parallel parking ability.