Adaptive Resonance Theory Neural Network Research Articles

Single particle mass spectral signatures from on-road and non-road vehicle exhaust particles and their application in refined source apportionment using deep learning

With advances in vehicle emission control technology, updating source profiles to meet the current requirements of source apportionment has become increasingly crucial. In this study, on-road and non-road vehicle particles were collected, and then the chemical compositions of individual particles were analyzed using single particle aerosol mass spectrometry. The data were grouped using an adaptive resonance theory neural network to identify signatures and establish a mass spectral database of mobile sources. In addition, a deep learning-based model (DeepAerosolClassifier) for classifying aerosol particles was established. The objective of this model was to accomplish source apportionment. During the training process, the model achieved an accuracy of 98.49 % for the validation set and an accuracy of 93.36 % for the testing set. Regarding the model interpretation, ideal spectra were generated using the model, verifying its accurate recognition of the characteristic patterns in the mass spectra. In a practical application, the model performed hourly source apportionment at three specific field monitoring sites. The effectiveness of the model in field measurement was validated by combining traffic flow and spatial information with the model results. Compared with other machine learning methods, our model achieved highly automated source apportionment while eliminating the need for feature selection, and it enables end-to-end operation. Thus, in the future, it can be applied in refined and online source apportionment of particulate matter.

Improving power cable partial discharge pattern recognition through gustafson-kessel fuzzy clustering techniques

Existing methods for recognizing partial discharge patterns in power cables do not utilize fuzzy clustering of the discharge signals, resulting in poor quality and low recall and precision of the pattern recognition. To address this, we propose a new approach for partial discharge pattern recognition in cables using Gustafson-Kessel(GK) Fuzzy Clustering. The method involves acquiring signals from a power cable partial discharge monitoring system and then processing the signals with GK fuzzy clustering. The clustered discharge signals are filtered with wavelet packet transforms before input into an improved adaptive resonance theory(ART) neural network for final pattern recognition. Experiments demonstrate the new technique achieves up to 98.7% recall and 85.6% precision for discharge pattern recognition, with discharge signal Signal Noise Ratio(SNR) between 55 dB and 62 dB and maximum recognition accuracy reaching 98%. The proposed fuzzy clustering-based pattern recognition approach significantly enhances partial discharge diagnostics for power cable monitoring.

Actuator-level motion and contact episode learning and classification using adaptive resonance theory

Several methods exist to detect and distinguish collisions of robotic systems with their environment, since this information is a critical dependency of many tasks. These methods are prevalently based on thresholds in combination with filters, models, or offline trained machine learning models. To improve the adaptation and thereby enable a more autonomous operation of robots in new environments, this work evaluates the applicability of an incremental learning approach. The method addresses online learning and recognition of motion and contact episodes of robotic systems from proprioceptive sensor data using machine learning. The objective is to learn new category templates representing previously encountered situations of the actuators and improve them based on newly gathered similar data. This is achieved using an artificial neural network based on adaptive resonance theory (ART). The input samples from the robot’s actuator measurements are preprocessed into frequency spectra. This enables the ART neural network to learn incrementally recurring episodic patterns from these preprocessed data. An evaluation based on preliminary experimental data from a grasping motion of a humanoid robot’s arm encountering contacts is presented and suggests that this is a promising approach.

Combining neural network and sequence analyses to determine animal call repertoires

Classifying frequency modulation patterns in animal signals is a challenging task depending on the level of stereotypy of call types. Not only do graded transitions between types create problems for defining them, but animals also vary single parameters within types without changing the overall shape of a modulation. One example for the latter is the variation in overall duration within call types of delphinids. Deecke et al. (1999) developed an adaptive resonance theory neural network for killer whale calls to standardize classification for dialect comparisons. This was later modified to make it usable for delphinid whistles by Deecke and Janik (2006) and is available as free software under the name ARTWARP. It creates categories by comparing frequency contours while time warping them to allow for variation in overall duration. Average frequency contours of all resulting call types are provided to describe repertoires once classification is complete. ARTWARP has recently been further improved to expand its analysis outputs and speed and can be used for the classification of any animal signals with a tonal component. Tests of ARTWARP in combination with the SIGID sequence analysis for signature whistles (Janik et al., 2013) provides an accurate repertoire of delphinid calls.

Tour Planning Design for Mobile Robots Using Pruned Adaptive Resonance Theory Networks

The development of intelligent algorithms for controlling autonom- ous mobile robots in real-time activities has increased dramatically in recent years. However, conventional intelligent algorithms currently fail to accurately predict unexpected obstacles involved in tour paths and thereby suffer from inefficient tour trajectories. The present study addresses these issues by proposing a potential field integrated pruned adaptive resonance theory (PPART) neural network for effectively managing the touring process of autonomous mobile robots in real-time. The proposed system is implemented using the AlphaBot platform, and the performance of the system is evaluated according to the obstacle prediction accuracy, path detection accuracy, time-lapse, tour length, and the overall accuracy of the system. The proposed system provide a very high obstacle prediction accuracy of 99.61%. Accordingly, the proposed tour planning design effectively predicts unexpected obstacles in the environment and thereby increases the overall efficiency of tour navigation.

A neural networks approach to predict call center calls of an internet service provider1

In today’s competitive business environment, companies are striving to reduce costs and workload of call centers while improving customer satisfaction. In this study, a framework is presented that predicts and encourages taking proactive actions to solve customer problems before they lead to a call to the call center. Machine learning techniques are implemented and models are trained with a dataset which is collected from an internet service provider’s systems in order to detect internet connection problems of the customers proactively. Firstly, two classification techniques which are multi perceptron neural networks and radial basis neural networks are applied as supervised techniques to classify whether the internet connection of customers is problematic or not. Then, by using unsupervised techniques, namely Kohonnen’s neural networks and Adaptive Resonance Theory neural networks, the same data set is clustered and the clusters are used for the customer problem prediction. The methods are then integrated with an ensemble technique bagging. Each method is implemented with bagging in order to obtain improvement on the estimation error and variation of the accuracy. Finally, the results of the methods applied for classification and clustering with and without bagging are evaluated with performance measures such as recall, accuracy and Davies-Bouldin index, respectively.

Role-Based Profiling Using Fuzzy Adaptive Resonance Theory for Securing Database Systems

Very large amounts of time and effort have been invested by the research community working on database security to achieve high assurance of security and privacy. An important component of a secure database system is intrusion detection system which has the ability to successfully detect anomalous behavior caused by applications and users. However, modeling the normal behavior of a large number of users in a huge organization is quite infeasible and inefficient. The main purpose of this research investigation is thus to model the behavior of roles instead of users by applying adaptive resonance theory neural network. The observed behavior which deviates from any of the established role profiles is treated as malicious. The proposed model has the advantage of identifying insider threat and is applicable for large organizations as it is based on role profiling instead of user profiling. The proposed system is capable of detecting intrusion with high accuracy along with minimized false alarms.

Size‐Resolved Mixing States and Sources of Amine‐Containing Particles in the East China Sea

AbstractSingle‐particle aerosol mass spectrometry was employed during a cruise campaign from 3 to 27 June 2017 to investigate the mixing states and sources of amine‐containing particles in the East China Sea. A total of 271,160 particles were successfully identified, and 19,861 of these particles contained amines. Monomethylamine (MMA), trimethylamine (TMA), and diethylamine (DEA) were the most abundant amines. They accounted for 50%, 16%, and 29%, respectively, of the amine‐containing particles. The size distributions (mean ± geometric standard deviation) of the MMA‐, TMA‐, and DEA‐containing particles peaked at 0.58 ± 0.12, 0.56 ± 0.10, and 0.40 ± 0.10 μm, respectively. Using an adaptive resonance theory neural network algorithm, four major clusters of amine‐containing particles were categorized: carbon‐rich particles, K‐rich particles, Na‐rich particles, and metal‐rich particles. The MMA‐containing particles contained abundant elemental carbon‐, K‐, Mn‐, and Fe‐rich particles, which exhibit higher concentrations in land air masses, indicating that the MMA mainly originated from terrestrial anthropogenic sources. The TMA‐containing particles contained abundant organic and elemental carbon‐, EC‐, K‐, and V‐rich particles. High concentrations of TMA‐containing particles were measured near ports, indicating the significant contribution of the ports. The DEA‐containing particles contained abundant organic and elemental carbon‐, K‐, Mn‐, Cu‐, and Na‐rich particles, and high concentrations of DEA‐containing particles were observed in remote marine areas, suggesting that both terrestrial anthropogenic and marine sources are important contributors to DEA‐containing particles.

Hardware Implementation of Parallelized Fuzzy Adaptive Resonance Theory Neural Network

A hardware implementation design of parallelized fuzzy Adaptive Resonance Theory neural network is described and simulated. Parallel category choice and resonance are implemented in the network. Continuous-time and discrete-time winner-take-all neural circuits identifying the largest of M inputs are used as the winner-take-all units. The continuous-time circuit is described by a state equation with a discontinuous right-hand side. The discrete-time counterpart is governed by a difference equation. Corresponding functional block-diagrams of the circuits include M feed-forward hard- limiting neurons and one feedback neuron, which is used to compute the dynamic shift of inputs. The circuits combine arbitrary finite resolution of inputs, high convergence speed to the winner-take-all operation, low computational and hardware implementation complexity, and independence of initial conditions. The circuits are also used for finding elements of input vector with minimal/maximal values to normalize them in the range [0,1].

Group incremental adaptive clustering based on neural network and rough set theory for crime report categorization

Explosively growing online text reports are mostly unstructured in nature. Many state-of-the-art techniques involving supervised, unsupervised or semi-supervised approaches have been developed in the recent years for automatic clustering of these reports. Annotation of online crime reports is a challenging task as various types of crime reports are frequently generated over time. To the best of the authors’ knowledge, this is the first attempt taken for group incremental adaptive clustering of crime reports integrating neural network and rough set theory. The proposed work initially identifies the named entities and selects only the context words within a pair of entities as a phrase. Thus every report is described by a collection of phrases. The phrases are vectorized using GloVe and a graph based clustering algorithm is applied to cluster all the collected phrases. The phrases within a cluster are considered as the similar type of phrases, called paraphrases and each report is represented by a binary vector of dimension equal to the number of clusters obtained. If a phrase of the report lies in a cluster then a ‘1’ is set at the corresponding position of the binary vector; otherwise it is set as ‘0’. Next, an adaptive resonance theory neural network is applied on the binary vector representation of the crime reports to generate a set of clusters of crime reports. When a new group of reports is available, the reports are transformed into binary form in the similar way and the rough set theory is applied on them. It puts many reports into existing clusters and for the remaining reports, adaptive resonance theory is further applied to modify the existing clusters and possibly generate the new clusters. Thus, in the dynamic environment when data are generated gradually over time, the proposed group incremental clustering algorithm is adapted to provide the updated set of clusters. The method has been applied on various crime report datasets and validated with the help of several cluster validation indices. The method is also compared with some state-of-the-art clustering algorithms to express its effectiveness and statistical significance in the domain of crime corpora.

A survey of adaptive resonance theory neural network models for engineering applications

This survey samples from the ever-growing family of adaptive resonance theory (ART) neural network models used to perform the three primary machine learning modalities, namely, unsupervised, supervised and reinforcement learning. It comprises a representative list from classic to contemporary ART models, thereby painting a general picture of the architectures developed by researchers over the past 30 years. The learning dynamics of these ART models are briefly described, and their distinctive characteristics such as code representation, long-term memory, and corresponding geometric interpretation are discussed. Useful engineering properties of ART (speed, configurability, explainability, parallelization and hardware implementation) are examined along with current challenges. Finally, a compilation of online software libraries is provided. It is expected that this overview will be helpful to new and seasoned ART researchers.

The newly described Araguaian river dolphins, Inia araguaiaensis (Cetartiodactyla, Iniidae), produce a diverse repertoire of acoustic signals.

The recent discovery of the Araguaian river dolphin (Inia araguaiaensis) highlights how little we know about the diversity and biology of river dolphins. In this study, we described the acoustic repertoire of this newly discovered species in concert with their behaviour. We analysed frequency contours of 727 signals (sampled at 10 ms temporal resolution). These contours were analyzed using an adaptive resonance theory neural network combined with dynamic time-warping (ARTwarp). Using a critical similarity value of 96%, frequency contours were categorized into 237 sound-types. The most common types were emitted when calves were present suggesting a key role in mother-calf communication. Our findings show that the acoustic repertoire of river dolphins is far from simple. Furthermore, the calls described here are similar in acoustic structure to those produced by social delphinids, such as orcas and pilot whales. Uncovering the context in which these signals are produced may help understand the social structure of this species and contribute to our understanding of the evolution of acoustic communication in whales.

Characterization of urban amine-containing particles in southwestern China: seasonal variation, source, and processing

Abstract. Amine-containing particles were characterized in an urban area of Chongqing during both summer and winter using a single-particle aerosol mass spectrometer (SPAMS). Among the collected particles, 12.7 % were amine-containing in winter and 8.3 % in summer. Amines were internally mixed with elemental carbon (EC), organic carbon (OC), sulfate, and nitrate. Diethylamine (DEA) was the most abundant among amine-containing particles. Wintertime amine-containing particles were mainly from the northwest direction where a forest park was located; in summer, they were from the northwest and southwest (traffic hub) directions. These origins suggest that vegetation and traffic were the primary sources of particulate amines. The average relative peak area of DEA depended strongly on humidity, indicating that the enhancement of DEA was possibly due to increasing aerosol water content and aerosol acidity. Using an adaptive resonance theory neural network (ART-2a) algorithm, four major types of amine-containing particles were clustered: amine–organic carbon (A-OC), A-OCEC, DEA-OC, and A-OCEC aged. The identified particle types implied that amines were taken up by particles produced from traffic and biomass burning. The knowledge gained in this study is useful to understand the atmospheric processing, origin, and sources of amine-containing particles in the urban area of Chongqing.

Hybrid Data Regression Model Based on the Generalized Adaptive Resonance Theory Neural Network

The Generalized Adaptive Resonance Theory (GART) model is a supervised online learning neural network based on an integration of Adaptive Resonance Theory (ART) and the Generalized Regression Neural Network (GRNN). It is capable of online learning, and is suitable for undertaking both classification and regression problems. In this paper, we further enhance GART (EGART) with four improvements to formulate a new EGART model. Three operating strategies for the EGART model to undertake regression problems are suggested. The first operating strategy is a fully online learning EGART model. The second operating strategy involves Ordering Algorithm for determining the presentation sequence of training samples during the initial training of EGART model. This strategy is considered as offline learning because a set of data samples must be available for the Ordering Algorithm to compute the best presentation sequence (hereinafter denoted as Ordered-EGART). The third operating strategy aims to demonstrate online learning capability of EART model (the first operating strategy) can still be resumed after training on the Ordered-EGART. It is most suitable for applications with a set of ready data samples and their sequences are predetermined by Ordering Algorithm prior to training of EGART model in offline mode, and triggers online learning when more new data samples become available (hereinafter denoted as IO-EGART). A series of experiments with five benchmark data sets from various application domains is conducted to assess and compare the effectiveness of the EGART model and three operating strategies with those of other methods published in literature as well as two fire safety engineering problems, i.e., predicting the thermal interface height in a single compartment fire and evacuation times in the event of fire. The results and comparisons with other approaches positively demonstrate the efficacy and applicability of EGART model as a useful data regression model for tackling fire safety engineering problems.

Socially Aware Caching in D2D Enabled Fog Radio Access Networks

As a promising architecture in the future networks, fog radio access network (F-RAN) can effectively improve network performance due to the advantages in both communication and computation aspects. In this paper, the data caching scheme is studied in a device-to-device (D2D) enabled the F-RAN from the social point of view. First, we help each access point (AP) to ascertain its best belonging community in terms of the content preference based on the projective adaptive resonance theory neural network (PART NN). The APs within identical community cache data cooperatively, which can reduce the caching redundancy. Considering the content preference of the AP may change dynamically, a community update scheme is also studied after the community construction is completed. In order to further reduce the content delivery delay, the D2D technology is resorted to and the most appropriate user equipment (UE) is selected to cache data for other UEs within each AP’s coverage. After that, an ant colony optimization scheme is proposed to find out the most requested contents needed to be cached by such most appropriate UE. The effectiveness of the proposed scheme is finally validated by the simulation experiments.

Multinodal load forecasting for distribution systems using a fuzzy-artmap neural network

This work proposes a predictor system (multinodal forecasting) considering several points of an electrical network, such as substations, transformers, and feeders, based on an adaptive resonance theory (ART) neural network family. It is a problem similar to global forecasting, with the main difference being the strategy to align the input and output of the data with several parallel neural modules. Considering that multinodal prediction is more complex compared to global prediction, the multinodal prediction will use a fuzzy-ARTMAP neural network and a global load participation factor. The advantages of this approach are as follows: (1) the processing time is equivalent to the processing required for global forecasting (i.e., the additional time processing is quite low); and (2) Fuzzy-ARTMAP neural networks converge significantly faster than backpropagation neural networks (improved benchmark in precision). The preference for neural networks of the ART family is due to the characteristic stability and plasticity that these architectures have to provide results in a fast and precise way. To test the proposed forecast system, the results are presented for nine substations from the database of an electrical company.

Behavioral similarity of residential customers using a neural network based on adaptive resonance theory

This article presents a methodology aiming at the comprehension and analysis of the residential electricity consumption habitual behavior by means of a similarity analysis, based on the use of an ART (Adaptive Resonance Theory) neural network, which is a neural network composed of two fuzzy ART modules whose training is performed in an unsupervised mode. ART neural networks are stable and plastic and these properties, combined with the processing of essentially binary data, give the neural system a wide capacity for producing objectives that may be easily modified to satisfy requirements predetermined by consumer. The expected result is to obtain information regarding the similarity of consumers. Thus, some benefits may be derived by consumers, such as improved habits of electricity consumption and better strategies for negotiating more favorable rates, especially in the case of smart grid systems. In this new electricity sector paradigm, there is a strong consumer trend for free choice among electricity suppliers. This methodology also benefits load forecasting studies at grid points, where there is greater uncertainty, e.g., the busbars that are closest to consumers, i.e., the uncertainties in the context of the total load forecasting system are increased from the global load to the final consumer.

A Fuzzy Adaptive Resonance Theory‐Based Model for Mix Proportion Estimation of High‐Performance Concrete

AbstractA new approach that adopts the use of fuzzy adaptive resonance theory (ART) neural network in estimating high‐performance concrete (HPC) mix proportion from experimental data is devised. The proposed model receives a set of desired concrete performances, searches for a set of mix proportions that is near to the desired concrete performances, classifies the mix proportions into clusters, measures the similarity between performances of deduced clusters with desired performances, and deduces a mix proportion. The proposed model was used to estimate the mix proportions of five batches of concrete based on the performance criteria of 7th and 28th day compressive strengths. The generated mix proportions were used in an experimental work and the errors were within 13% for 7th compressive strength; and 7% for the 28th day compressive strength, signifying the reliability of the fuzzy ART‐based model in estimating the mix proportion of HPC. This article contributes to an alternative method of mix proportion estimation of HPC by avoiding the use of complicated function approximation techniques.

Based on Soft Competition ART Neural Network Ensemble and Its Application to the Fault Diagnosis of Bearing

This paper presents a novel method for fault diagnosis based on an improved adaptive resonance theory (ART) neural network and ensemble technique. The method consists of three stages. Firstly, the improved ART neural network is comprised of the soft competition technique based on fuzzy competitive learning (FCL) and ART based on Yu’s norm, the neural nodes in the competition layer are trained according to the degree of membership between the mode node and the input, and then fault samples are classified in turn. Secondly, with the distance evaluation technique, the optimal features are obtained from the statistical characteristics of original signals and wavelet coefficients. Finally, the optimal features are input into the neural network ensemble (NNE) based on voting method to identify the different fault categories. The proposed method is applied to the fault diagnosis of rolling element bearings, and testing results show that the neural network ensemble can reliably classify different fault categories and the degree of faults, which has a better classification performance compared with the single neural network.

DYNAMIC AND INCREMENTAL EXPLORATION STRATEGY IN FUSION ADAPTIVE RESONANCE THEORY FOR ONLINE REINFORCEMENT LEARNING

One of the fundamental challenges in reinforcement learning is to setup a proper balance between exploration and exploitation to obtain the maximum cummulative reward in the long run. Most protocols for exploration bound the overall values to a convergent level of performance. If new knowledge is inserted or the environment is suddenly changed, the issue becomes more intricate as the exploration must compromise the pre-existing knowledge. This paper presents a type of multi-channel adaptive resonance theory (ART) neural network model called fusion ART which serves as a fuzzy approximator for reinforcement learning with inherent features that can regulate the exploration strategy. This intrinsic regulation is driven by the condition of the knowledge learnt so far by the agent. The model offers a stable but incremental reinforcement learning that can involve prior rules as bootstrap knowledge for guiding the agent to select the right action. Experiments in obstacle avoidance and navigation tasks demonstrate that in the configuration of learning wherein the agent learns from scratch, the inherent exploration model in fusion ART model is comparable to the basic E-greedy policy. On the other hand, the model is demonstrated to deal with prior knowledge and strike a balance between exploration and exploitation.