This Letter presents meaningful results that demonstrate the reduction of dimensionality by spiking neural networks (SNNs) on benchmarking data. This experimental scheme includes metaheuristics, namely, the artificial bee colony algorithm (ABC algorithm) for finding optimal conductance values in the SNNs. Therefore, the objective function in the used ABC algorithm leads the SNNs to compute the principal component analysis (PCA), efficiently. The eigendecomposition of the information drawn by the SNNs in the training phase is the base of the formulated objective function. In these experiments, the Izhikevich model represents the spiking neurons, which have biological plausibility with parameters for reproducing a uniform firing rate. The visualisation of clusters in the 3D PCA space, whose sample values are compared with the PCA function in Matlab, is also shown; this comparison demonstrates an acceptable error in the MSE sense.

This paper presents the performance of optimal filtering; LMS with a nonvolatile analog memory cell fabricated through 1.2um CMOS process, for the adaptive identification signal problem, where transfer functions are unknown and changing. The memory stores the weight in the filter as charge on the floating gate of a transistor pMOS. The update is linear, using a pulse density modulation scheme by means of tunneling and injection mechanisms. The LMS algorithm is implemented digitally off chip, and it does not require the signal to be piecewise stationary, and requires no manual operation other than selection of the step-size.

This work deals with the presentation of a spiking neural network as a means for efficiently solving the reduction of dimensionality of data in a nonlinear manner. The underneath neural model, which can be integrated as neuromorphic hardware, becomes suitable for intelligent processing in edge computing within Internet of Things systems. In this sense, to achieve a meaningful performance with a low complexity one-layer spiking neural network, the training phase uses the metaheuristic Artificial Bee Colony algorithm with an objective function from the principals in the machine learning science, namely, the modified Stochastic Neighbor Embedding algorithm. To demonstrate this fact, complex benchmark data were used and the results were compared with those generated by a reference network with continuous-sigmoid neurons. The goal of this work is to demonstrate via numerical experiments another method for training spiking neural networks, where the used optimizer comes from metaheuristics. Therefore, the key issue is defining the objective function, which can relate optimally the information at both sides of the spiking neural network. Certainly, machine learning techniques have advanced in defining efficient loss functions that can become suitable objective function candidates in the metaheuristic training phase. The practicality of these ideas is shown in this article. We use MSE values for evaluating the relative quality of the results and also co-ranking matrices.

The ubiquitous classification task is a computing process within current machine learning frameworks, which are based on neural networks. These neural systems are trained using datasets created from technical, scientific or engineering applications, with excellent outcomes. This is because the associated training algorithms are variants of the gradient descending optimization procedure e.g. Adam or Adamax, that cope with local minima; however, they are computationally complex. The Extreme Learning Machine (ELM) neural model relaxes this numerical issue by computing Moore-Penrose matrices as a training method with acceptable performance values. In this context, Field Programmable Gate Array (FPGA) accelerators have been developed for this type of matrices, which include special-purpose cores. Considering the present work, it introduces the standard Widrow-Hoff neural training rule as the key process to compute efficiently Moore-Penrose matrices in classification tasks using the ELM model. Moreover, the use of the vectorial form instead of the matrixial representation for this training rule reduces the hardware resources in the used FPGA, leading to a low-cost system, whose performance in terms of time and accuracy is suitable for moderate matrix sizes in diverse classification problems. To prove this, six datasets from real-world applications were used, i.e. diabetes, cancer, sonar, ionosphere, wine and vehicle, which serve very often as benchmarks in machine learning research. Therefore, the innovation of this work is the use of a neural learning rule for computing Moore-Penrose matrices with a generic FPGA, with three meaningful features as follows. (1) this technique reduces substantially the usage of numerical resources in the FPGA, (2) the used hardware framework is based on standard development prototyping boards, reducing cost and, (3) although the learning algorithm in this FPGA trains in a stochastic manner, the chosen learning rate reduces this effect without affecting convergence velocity.

In this paper, a Least Mean Square (LMS) programming scheme is used to set the offset voltage of two operational amplifiers that were built using floating-gate transistors, enabling a 0.95 V_RMS trimmer-less flame detection sensor. The programming scheme is capable of setting the offset voltage over a wide range of values by means of electron injection. The flame detection sensor consists of two programmable offset operational amplifiers; the first amplifier serves as a 26 μV offset voltage follower, whereas the second amplifier acts as a programmable trimmer-less voltage comparator. Both amplifiers form the proposed sensor, whose principle of functionality is based on the detection of the electrical changes produced by the flame ionization. The experimental results show that it is possible to measure the presence of a flame accurately after programming the amplifiers with a maximum of 35 LMS-algorithm iterations. Current commercial flame detectors are mainly used in absorption refrigerators and large industrial gas heaters, where a high voltage AC source and several mechanical trimmings are used in order to accurately measure the presence of the flame.

Spiking neural networks (SNNs) are numerical models whose computational properties come from interconnected biological neurons and synapses, leading to the development of neuromorphic systems in either software or hardware. These models in digital hardware operate in the discrete time domain, computing mainly the membrane potential of a neuron, where the arithmetic processes can be reduced to additions with minimal loss in accuracy. This advantage comes from approximating the solution of the original differential membrane equation by a recurrent discrete-time algorithm. In this work, we present a discrete-time SNN design using machine learning tools within the PyTorch framework. Its realization with a Field Programmable Gate Array (FPGA) can efficiently classify industrial gases, which are dangerous. It is demonstrated using available datasets created by experimental measurement boards with metal oxide gas sensors that show variability. The SNN model in this work was trained with the Adam optimizer and a weight clipping procedure to enhance convergence, where the classification efficiency was achieved at 95.7%. Then, the synaptic parameters were exported to an optimized FPGA. Three meaningful contributions of this digital neuromorphic system are as follows: (1) the proposed SNN approximates continuous membrane potentials in neurons with simple time series processing modules, (2) the neuron model, which is the Leaky Integrate and Fire (LIF) model with threshold capability, is parallelizable in hardware, and (3) the final SNN hardware prototype demonstrates its performance in terms of energy efficiency in a classification task scheme. Moreover, the final values of the weighting parameters between neurons are well-suited to an analog spiking system, serving as a hardware counterpart to the digital one in this work.