Search

Coming up with a model which matches biological observations more closely has always been one of the main challenges in the field of artificial neural networks. Lately, an ionic model of reservoir networks containing spiking neurons (ILS-based reservoir network) has been proposed which seems to replicate some of the biological processes we have observed up until now. This paper presents a local learning rule for the ILS-based reservoir inspired by the biological fact that each incoming stimulus causes the formation of new dendritic spines, producing new synapses. This property may result in a higher degree of neuroplasticity, leading to a higher learning capacity. To evaluate the proposed learning rule, which somehow takes the structural plasticity into account, several experiments have been designed that show its stability and its effectiveness on separation property, classification accuracy, and reliability. Several benchmark cases have also been discussed and their results through separation and classification evaluation metrics, demonstrate the superior performance of the ILS-based reservoir network trained by the proposed learning rule compared to the untrained similar network. [ABSTRACT FROM AUTHOR]

Active Learning Method (ALM) is one of the powerful tools in soft computing and it is inspired by the human brain capabilities in approaching complicated problems. ALM, which is in essence an adaptive fuzzy learning algorithm, tries to model a Multi-Input Single-Output system with several single-input single-output subsystems. Each of these subsystems is then modeled by an ink drop spread (IDS) plane. IDS operator, which is the main processing engine of ALM, extracts two kinds of informative features, Narrow Path and Spread, from each IDS plane without complicated computations. These features from all IDS planes are then aggregated in the inference engine. Despite the great performance of ALM in different applications, an efficient hardware implementation has remained a challenge, which is mainly due to considerably high memory requirement of IDS operation. In this paper, in a novel approach to IDS operation, we propose an abstract representation of the IDS planes which minimizes the memory requirement and the computational cost, and consequently, benefits the hardware implementation in terms of area and speed. The proposed approach is fully compatible with memristor-crossbar implementation with an adaptive learning capability. Simpler learning algorithm and higher speed make our proposed algorithm suitable for applications where real-time process, low-cost and small implementation are of high priority. Applications in the classification of real-world datasets and function approximation are provided to confirm the effectiveness of the algorithm. Eventually, the paper concludes that the proposed computing structure provides a synergy between artificial neural networks and fuzzy domains. [ABSTRACT FROM AUTHOR]

In this paper the Unsupervised Active Learning Method (UALM), a novel clustering method based on the Active Learning Method (ALM) is introduced. ALM is an adaptive recursive fuzzy learning algorithm inspired by some behavioral features of human brain functionality. UALM is a density-based clustering algorithm that relies on discovering densely connected components of data, where it can find clusters of arbitrary shapes. This approach is a noise-robust clustering method. The algorithm first blurs the data points as ink drop patterns, then summarizes the effects of all data points, and finally puts a threshold on the resulting pattern. It uses the connected-component algorithm for finding clusters. Then determines cluster centers by intersecting the narrow-paths. Experimental results confirmed the superiority of our proposed method compared to the two most well-known density-based clustering algorithms, DBSCAN and DENCLUE. [ABSTRACT FROM AUTHOR]

Active Learning Method (ALM) is a powerful fuzzy soft computing tool, developed originally in order to promote an engineering realization of human brain. This algorithm, as a macro-level brain imitation, has been inspired by some behavioral specifications of human brain and active learning ability. ALM is an adaptive recursive fuzzy learning algorithm, in which a complex Multi Input, Multi Output system can be represented as a fuzzy combination of several Single-Input, Single-Output systems. SISO systems as associative layer of algorithm capture partial spatial knowledge of sample data space, and enable a granular knowledge resolution tuning mechanism through the learning process. The knowledge in each sub-system and its effectiveness in the whole system would be extracted by Ink Drop Spread in brief IDS operator and consolidated using a Fuzzy Rule Base (FRB), in order to acquire expert knowledge. In this paper we investigate ALM as a conspicuous classifier in different types of classification problems. Also, a new ALM architecture to actively analyze ill-balanced image patterns is proposed. Different types of data sets are used as a benchmark, including a remote sensing image classification problem, to evaluate the ALM Classifier (ALMC). With active pattern generation ability and knowledge resolution tuning, ALMC has been distinguished from many conventional classification tools especially for complex structures and image patterns analysis. This work demonstrates that ALMC is a good noise robust and active classifier, which is adaptively adjusted through structural evolution and pattern evaluation mechanism. These remarkable capabilities, along with its straightforward learning process, make ALMC as a convenient soft computing tool to use in different types of low dimensional pattern recognition problems. [ABSTRACT FROM AUTHOR]

Abstract: Constrained clustering methods (that usually use must-link and/or cannot-link constraints) have been received much attention in the last decade. Recently, kernel adaptation or kernel learning has been considered as a powerful approach for constrained clustering. However, these methods usually either allow only special forms of kernels or learn non-parametric kernel matrices and scale very poorly. Therefore, they either learn a metric that has low flexibility or are applicable only on small data sets due to their high computational complexity. In this paper, we propose a more efficient non-linear metric learning method that learns a low-rank kernel matrix from must-link and cannot-link constraints and the topological structure of data. We formulate the proposed method as a trace ratio optimization problem and learn appropriate distance metrics through finding optimal low-rank kernel matrices. We solve the proposed optimization problem much more efficiently than SDP solvers. Additionally, we show that the spectral clustering methods can be considered as a special form of low-rank kernel learning methods. Extensive experiments have demonstrated the superiority of the proposed method compared to recently introduced kernel learning methods. [Copyright &y& Elsevier]

Metric learning is a powerful approach for semi-supervised clustering. In this paper, a metric learning method considering both pairwise constraints and the geometrical structure of data is introduced for semi-supervised clustering. At first, a smooth metric is found (based on an optimization problem) using positive constraints as supervisory information. Then, an extension of this method employing both positive and negative constraints is introduced. As opposed to the existing methods, the extended method has the capability of considering both positive and negative constraints while considering the topological structure of data. The proposed metric learning method can improve performance of semi-supervised clustering algorithms. Experimental results on real-world data sets show the effectiveness of this method. [ABSTRACT FROM AUTHOR]

Recombination in Genetic Algorithms (GA) is supposed to extract the component characteristics from two parents and reassemble them in different combinations, hopefully producing an offspring that has the good characteristics of both parents, and this requires explicit chromosome and recombination, operator by design. This paper presents a novel evolutionary approach based on symbiogenesis which uses symbiotic combination instead of sexual recombination, and by using this operator, it requires no domain knowledge for chromosome or combination operator design. The algorithm is benchmarked on three problem sets: combinatorial optimization category, deceptive problems, and fully deceptive problems. The results, compared with that of standard genetic algorithm and symbiotic evolutionary adaptation model, show higher success rates and faster results. [ABSTRACT FROM AUTHOR]

Evolutionary Algorithms are vastly used in development of rule based classifier systems in data mining where the rule base is usually a set of If-Then rules and an evolutionary trait develops and optimizes these rules. Genetic Algorithm is usually a favorite solution for such tasks as it globally searches for good rule-sets without any prior bias or greedy force, but it is usually slow. Also, designing a good genetic algorithm for rule base evolution requires the design of a recombination operator that merges two rule bases without disrupting the functionalities of each of them. To overcome the speed problem and the need to design recombination operator, this paper presents a novel algorithm for rule base evolution based on natural process of symbiogenesis. The algorithm uses symbiotic combination operator instead of traditional sexual recombination operator of genetic algorithms. This operator takes two chromosomes with different number of genes (rules here) and merges them by combining all the information content of both chromosomes. Using this operator results in two major advantages: First, it totally removes the need to design the recombination operator and therefore is easier to use; second, it outperforms traditional genetic algorithm both in emergence speed and classification rate, this is tested and presented on some globally used benchmarks. [ABSTRACT FROM AUTHOR]

Persian is a fully cursive handwriting in which each character may take different forms in different parts of the word, characters overlap and there is a wide range of possible styles. These complexities make automatic recognition of Persian a very hard task. This paper presents a novel approach on recognition of such writings systems which is based on the description of input stream by a sequence of fuzzy linguistic terms; representation of character patterns with the same descriptive language; and comparison of inputs with character patterns using a novel elastic pattern matching approach. As there is no general benchmark for recognition of Persian handwriting, the approach has been tested on the set of words in first primary Iranian school books including 1250 words resulting in 78% correct recognition without dictionary and 96% with dictionary. [ABSTRACT FROM AUTHOR]

Abstract: Fuzzy logic is a suitable approach for linguistic knowledge representation and Artificial Life (A-Life) is a field of study dedicated to studying life and synthesis of creatures like those of real world. Thus, as our understanding about life is more in linguistic form than exact numbers, fuzzy logic seems to be a suitable logic for artificial creatures. In this paper, we have tried to apply fuzzy logic to an A-life environment and have studied the changes in capabilities of its creatures when using fuzzy minds. To do this, we have selected an A-life model called Zamin, which has creatures with case-based reasoner brains who must learn their environmental rules, search for food, avoid predators and keep up their species. Then we have replaced their brains with fuzzy self-organizing controllers and have compared the results of this replacement with former creatures. The comparisons are done both in separate environments, where measures of successful life are recorded and compared, and in common arenas, where they need to compete for resources and escape predators. [Copyright &y& Elsevier]

As technology size scales down, magnetic tunnel junctions (MTJs) as a promising technology are becoming more and more sensitive to process variation, especially in oxide barrier thickness. Process variation particularly affects the cell resistance and the critical switching current for the smaller dimensions. This article proposes an MTJ cell with one free and two pinned layers, which highly improves the process variation robustness. By employing the spin transfer torque (STT)-spin-Hall effect (SHE) switching method, our proposed MTJ cell improves the switching speed and lowers the switching power consumption. Per simulations, an MRAM cell built with the proposed MTJ cell offers up to 36% lower total power consumption, up to 32% higher write performance, and 35% higher read performance over the previous state-of-the-art MRAMs. Also, the error rate resulting from process variation is 1.4%, which is about one-tenth of the best examples reported in the literature. [ABSTRACT FROM AUTHOR]

In this paper, introducing a new hardware concept which has the ability of FUZZY computation in a non-exact naturally FUZZY method, outlines of human brain simulation in hardware level will be presented. It will be shown that a soft computer may be implemented by using unique hardware structures for processor and memory units. Considering the learning procedure of human being and simulating it, using the introduced soft computer, a total energy decreasing trend will be observed in the soft computer structure. The generalization of this trend will be regarded as an equivalent concept for human intelligence. Therefore, complicated concepts such as human intelligence and also imagination and inference procedures may appear as natural abilities of the proposed soft computer. It also will be shown, when starting with observing a phenomenon, the whole procedure, from modeling to controller designing, including the recognition of goal function and effective inputs, may be done automatically and in an intelligent way. [ABSTRACT FROM AUTHOR]

Market prediction has been an important machine learning research topic in recent decades. A neglected issue in prediction is having a model that can simultaneously pay attention to the interaction of global markets along historical data of the target markets being predicted. As a solution, we present a hybrid supervised semi-supervised model called HyS3 for direction of movement prediction. The graph-based semi-supervised part of HyS3 models the markets global interactions through a network designed with a novel continuous Kruskal-based graph construction algorithm called ConKruG. The supervised part of the model injects results extracted from each market’s historical data to the network whenever the hybrid model allows with an innovative conditional mechanism. The significance of higher prediction accuracy of HyS3 is comparing to other models is proved statistically against other models including supervised models and network-based semi-supervised predictions. [ABSTRACT FROM AUTHOR]

Internet traffic recognition is essential for access providers since it helps them define adapted priorities in order to enhance user experience, e.g., a high priority for an audio conference and a low priority for a file transfer. As internet traffic becomes increasingly encrypted, the main classic traffic recognition technique, payload inspection, is rendered ineffective. Hence this paper uses machine learning techniques looking only at packet size and time of arrival. For the first time, Spiking neural networks (SNNs), which are inspired by biological neurons, were used for this task for two reasons. Firstly, they can recognize time-related data packet features. Secondly, they can be implemented efficiently on neuromorphic hardware. Here we used a simple feedforward SNN, with only one fully connected hidden layer, and trained in a supervised manner using the new method known as Surrogate Gradient Learning. Surprisingly, such a simple SNN reached an accuracy of 95.9% on ISCX datasets, outperforming previous approaches. Besides better accuracy, there is also a significant improvement in simplicity: input size, the number of neurons, trainable parameters are all reduced by one to four orders of magnitude. Next, we analyzed the reasons for this good performance. It turns out that, beyond spatial (i.e., packet size) features, the SNN also exploits temporal ones, mainly the nearly synchronous (i.e., within a 200 ms range) arrival times of packets with specific sizes. Taken together, these results show that SNNs are an excellent fit for encrypted internet traffic classification: they can be more accurate than conventional artificial neural networks (ANN), and they could be implemented efficiently on low-power embedded systems. [ABSTRACT FROM AUTHOR]

Market prediction has been an important research problem for decades. Having better predictive models that are both more accurate and faster has been attractive for both researchers and traders. Among many approaches, semi-supervised graph-based prediction has been used as a solution in recent researches. Based on this approach, we present two prediction models. In the first model, a new network structure is introduced that can capture more information about markets' direction of movements compared to the previous state of the art methods. Based on this novel network, a new algorithm for semi-supervised label propagation is designed that is able to prediction the direction of movement faster and more accurately. The second model is a mixture of experts system that decides between supervised or semi-supervised approaches. Besides this, the model gives us the ability to identify the markets that their data are helpful in constructing the network. Our models are shown to be both faster regarding computational complexity and running time and more accurate in prediction comparing to best rival models in literature of graph-based semi-supervised prediction. The results are also tested to be statistically significant. • A new semi-supervised graph-based classification method is presented. • A new way of interaction graph construction in time series datasets is proposed. • A novel graph edge-weighting method in semi-supervised learning is presented. • A new mixture of experts system for model selection and graph-design is presented. [ABSTRACT FROM AUTHOR]

Preisach model is a well-known hysteresis identification method in which the hysteresis is modeled by linear combination of hysteresis operators. Although Preisach model describes the main features of system with hysteresis behavior, due to its rigorous numerical nature, it is not convenient to use in real-time control applications. Here a novel neural network approach based on the Preisach model is addressed, provides accurate hysteresis nonlinearity modeling in comparison with the classical Preisach model and can be used for many applications such as hysteresis nonlinearity control and identification in SMA and Piezo actuators and performance evaluation in some physical systems such as magnetic materials. To evaluate the proposed approach, an experimental apparatus consisting one-dimensional flexible aluminum beam actuated with an SMA wire is used. It is shown that the proposed ANN-based Preisach model can identify hysteresis nonlinearity more accurately than the classical one. It also has powerful ability to precisely predict the higher-order hysteresis minor loops behavior even though only the first-order reversal data are in use. It is also shown that to get the same precise results in the classical Preisach model, many more data should be used, and this directly increases the experimental cost. [ABSTRACT FROM AUTHOR]