The inability to trace an AI’s reasoning process and understand why it makes each decision is known as the black box problem. This remains one of the major barriers to the trusted and widespread use of machine learning in many application domains. The paper explores pattern recognition performance and learning dynamics of the Tsetlin Machine – a new explainable logic-based machine-learning approach. Tsetlin Machine uses a collection of finite-state automata with a unique logic-based learning mechanism and provides a promising alternative to Artificial Neural Networks with several advantages, such as interpretability, low complexity, suitability for hardware implementation and high performance. This work investigates Tsetlin Machine’s mechanism for constructing conjunctive clauses from data and their interpretation for pattern recognition on several datasets. We demonstrate that during training the logical clauses learn persistent sub-patterns within the class. Each clause creates a class template by clustering a certain number of similar class samples, combining them through literal-wise logical conjunction (i.e., AND-ing). The number of class samples that each clause combines depends on Tsetlin Machine’s hyperparameters. The more class samples that are combined, the more general the clauses become. The paper aims at uncovering how Tsetlin Machine’s hyperparameters influence the balance between clause generalization and specialization and how this affects the accuracy of pattern recognition. It also studies the evolution of the machine’s internal state, its convergence and training completion.

We present an all-digital programmable machine learning accelerator chip for image classification, underpinning on the Tsetlin machine (TM) principles. The TM is an emerging machine learning algorithm founded on propositional logic, utilizing sub-pattern recognition expressions called clauses. The accelerator implements the coalesced TM version with convolution, and classifies booleanized images of$28\times 28$pixels with 10 categories. A configuration with 128 clauses is used in a highly parallel architecture. Fast clause evaluation is achieved by keeping all clause weights and Tsetlin automata (TA) action signals in registers. The chip is implemented in a 65 nm low-leakage CMOS technology, and occupies an active area of 2.7 mm2. At a clock frequency of 27.8 MHz, the accelerator achieves 60.3 k classifications per second, and consumes 8.6 nJ per classification. This demonstrates the energy-efficiency of the TM, which was the main motivation for developing this chip. The latency for classifying a single image is$25.4~\mu $s which includes system timing overhead. The accelerator achieves 97.42%, 84.54% and 82.55% test accuracies for the datasets MNIST, Fashion-MNIST and Kuzushiji-MNIST, respectively, matching the TM software models.

Recent advancements in artificial intelligence hold ample potential for monitoring applications using surveillance cameras. However, concerns about privacy and model bias have made it challenging to utilize them in public. Although de-identification approaches have been proposed in the literature, aiming to achieve a certain level of anonymization (AN), most of them employ deep learning models that are computationally demanding for real-time edge deployment. This study revisits conventional AN solutions for privacy protection and real-time video anomaly detection (VAD) applications. We propose a lightweight adaptive AN for VAD (LA3D) that employs dynamic adjustment to enhance full-body privacy protection. We have evaluated privacy protection and VAD utility retention efficacy using several publicly available datasets to examine the strengths and weaknesses of different AN methods and highlight the promising leverage of our approach. Our experiment demonstrates that the LA3D enables substantial improvement in privacy AN without severely degrading VAD efficacy, outperforming conventional and deep learning approaches. Code: https://github.com/muleina/LA3D.

Tsetlin Machines (TMs) have garnered increasing interest for their ability to learn concepts via propositional formulas and their proven efficiency across various application domains. Despite this, the convergence proof for the TMs, particularly for the AND operator (conjunction of literals), in the generalized case (inputs greater than two bits) remains an open problem. This paper aims to fill this gap by presenting a comprehensive convergence analysis of Tsetlin automaton-based Machine Learning algorithms. We introduce a novel framework, referred to as Probabilistic Concept Learning (PCL), which simplifies the TM structure while incorporating dedicated feedback mechanisms and dedicated inclusion/exclusion probabilities for literals. Given n features, PCL aims to learn a set of conjunction clauses Ci each associated with a distinct inclusion probability pi. Most importantly, we establish a theoretical proof confirming that, for any clause k, PCL converges to a conjunction of literals when pk is between 0.5 and 1. This result serves as a stepping stone for future research on the convergence properties of Tsetlin automaton-based learning algorithms. Our findings not only contribute to the theoretical understanding of Tsetlin automaton-based learning algorithms but also have implications for their practical application, potentially leading to more robust and interpretable machine learning models.

global navigation satellite systems (GNSSs) are vulnerable to jamming, which can degrade or even disable their services by interfering with signal reception. Such interference may interrupt GNSS positioning, navigation, timing, and communication functions. Therefore, robust jamming detection strategies are essential to ensure continuous and reliable services. Many existing jamming detection approaches use closed box machine learning (ML) models, which lack transparency and accuracy. In this article, we propose a novel mixture of convolutional Tsetlin machine experts (MixCTME) approach, using diverse spectrogram visualizations to enhance the accuracy and interpretability of jamming signal classification. To facilitate this, we collected raw in-phase and quadrature (IQ) data and named it the RealRFI dataset, which corresponds to ten different jamming classes. The data were collected through multiple radio-frequency interference (RFI) monitoring stations deployed at SINTEF companies across Europe and Scandinavia. Next, diverse visualizations were generated using short time Fourier transform (STFT), dominant frequency, fast Fourier transform (FFT), power spectral density, and standard deviation. These visualizations were binarized using the Otsu thresholding technique, and the binarized spectrograms were fed to the MixCTME model. Our method utilizes a combination of distinct convolutional Tsetlin machine (CTM) experts and employs a novel nonlinear technique, referred to as confidence-based gating, for weighting the experts’ inputs to make the final decision. We compared our method against state-of-the-art approaches via the self-collected and labeled dataset. Through experiments, our model achieved an accuracy of 90.70% on the collected data and 99.46% on a benchmark dataset, outperforming the state-of-the-art approaches. Additionally, we highlight the trustworthiness, scalability, and interpretability of the proposed approach, presenting a promising solution for accurate and interpretable jamming classification.

The Tsetlin Machine (TM) is a novel machine learning algorithm that uses Tsetlin automata (TAs) to define propositional logic expressions (clauses) for classification. This paper describes a field-programmable gate array (FPGA) accelerator for image classification based on the Convolutional Coalesced Tsetlin Machine. The accelerator classifies booleanized images of$28\times 28$pixels into 10 classes, and is configured with 128 clauses in a highly parallel architecture. To achieve fast clause evaluation and class prediction, the TA action signals and the clause weights per class are available from registers. Full on-device training is included, and the TAs are implemented with 34 Block RAM (BRAM) instances which operate in parallel. Each BRAM is addressed by the clause number and has a 72-bit word width that supports 8 TAs. The design is implemented in a Xilinx Zynq Ultrascale+ XCZU7 FPGA. Running at 50 MHz, the accelerator core achieves 134k image classifications per second, with an energy consumption per classification of$13.3~\mu $J. A single training epoch of 60k samples requires a processing time of 1.5 seconds. The accelerator obtains a test accuracy of 97.6% on MNIST, 84.1% on Fashion-MNIST and 82.8% on Kuzushiji-MNIST.

Since the early 1960s, the paradigm of learning automata (LA) has experienced abundant interest. Arguably, it has also served as the foundation for the phenomenon and field of reinforcement learning (RL). Over the decades, new concepts and fundamental principles have been introduced to increase the LA's speed and accuracy. These include using probability updating functions, discretizing the probability space, and using the "Pursuit" concept. Very recently, the concept of incorporating "structure" into the ordering of the LA's actions has improved both the speed and accuracy of the corresponding hierarchical machines, when the number of actions is large. This has led to the ϵ -optimal hierarchical continuous pursuit LA (HCPA). This article pioneers the inclusion of all the above-mentioned phenomena into a new single LA, leading to the novel hierarchical discretized pursuit LA (HDPA). Indeed, although the previously proposed HCPA is powerful, its speed has an impediment when any action probability is close to unity, because the updates of the components of the probability vector are correspondingly smaller when any action probability becomes closer to unity. We propose here, the novel HDPA, where we infuse the phenomenon of discretization into the action probability vector's updating functionality, and which is invoked recursively at every stage of the machine's hierarchical structure. This discretized functionality does not possess the same impediment, because discretization prohibits it. We demonstrate the HDPA's robustness and validity by formally proving the ϵ -optimality by utilizing the moderation property. We also invoke the submartingale characteristic at every level, to prove that the action probability of the optimal action converges to unity as time goes to infinity. Apart from the new machine being ϵ -optimal, the numerical results demonstrate that the number of iterations required for convergence is significantly reduced for the HDPA, when compared to the state-of-the-art HCPA scheme.

Logic-based machine learning has the crucial advantage of transparency. However, despite significant recent progress, further research is needed to close the accuracy gap between logic-based architectures and deep neural network ones. This paper introduces a novel variant of the Tsetlin machine (TM) that randomly drops clauses, the logical learning element of TMs. In effect, TM with Drop Clause ignores a random selection of the clauses in each epoch, selected according to a predefined probability. In this way, the TM learning phase becomes more diverse. To explore the effects that Drop Clause has on accuracy, training time and robustness, we conduct extensive experiments on nine benchmark datasets in natural language processing (IMDb, R8, R52, MR, and TREC) and image classification (MNIST, Fashion MNIST, CIFAR-10, and CIFAR-100). Our proposed model outperforms baseline machine learning algorithms by a wide margin and achieves competitive performance compared with recent deep learning models, such as BERT-Large and AlexNet-DFA. In brief, we observe up to +10% increase in accuracy and 2x to 4x faster learning than for the standard TM. We visualize the patterns learnt by Drop Clause TM in the form of heatmaps and show evidence of the ability of drop clause to learn more unique and discriminative patterns. We finally evaluate how Drop Clause affects learning robustness by introducing corruptions and alterations in the image/language test data, which exposes increased learning robustness.

Tsetlin Machine (TM) is a logic-based machine learning approach with the crucial advantages of being transparent and hardware-friendly. While TMs match or surpass deep learning accuracy for an increasing number of applications, large clause pools tend to produce clauses with many literals (long clauses). As such, they become less interpretable. Further, longer clauses increase the switching activity of the clause logic in hardware, consuming more power. This paper introduces a novel variant of TM learning -- Clause Size Constrained TMs (CSC-TMs) -- where one can set a soft constraint on the clause size. As soon as a clause includes more literals than the constraint allows, it starts expelling literals. Accordingly, oversized clauses only appear transiently. To evaluate CSC-TM, we conduct classification, clustering, and regression experiments on tabular data, natural language text, images, and board games. Our results show that CSC-TM maintains accuracy with up to 80 times fewer literals. Indeed, the accuracy increases with shorter clauses for TREC and BBC Sports. After the accuracy peaks, it drops gracefully as the clause size approaches one literal. We finally analyze CSC-TM power consumption and derive new convergence properties.

Abstract Due to the high arithmetic complexity and scalability challenges of deep learning, there is a critical need to shift research focus towards energy efficiency. Tsetlin Machines (TMs) are a recent approach to machine learning (ML) that has demonstrated significantly reduced energy compared to neural networks alike, while providing comparable accuracy on several benchmarks. However, TMs rely heavily on energy‐costly random number generation to stochastically guide a team of Tsetlin Automata (TA) in TM learning. In this paper, we propose a novel finite‐state learning automaton that can replace the TA in the TM, for increased determinism. The new automaton uses multi‐step deterministic state jumps to reinforce sub‐patterns, without resorting to randomization. A determinism parameter finely controls trading off the energy consumption of random number generation, against randomization for increased accuracy. Randomization is controlled by flipping a coin before every state jump, ignoring the state jump on tails. For example, makes every update random and makes the automaton completely deterministic. Both theoretically and empirically, we establish that the proposed automaton converges to the optimal action almost surely. Further, used together with the TM, only substantial degrees of determinism reduce accuracy. Energy‐wise, random number generation constitutes switching energy consumption of the TM, saving up to 11 mW power for larger datasets with high values. Our new learning automaton approach thus facilitates low‐energy ML.

The Tsetlin Machine (TM) is a novel machine learning algorithm with several distinct properties, including transparent inference and learning using hardware-near building blocks. Although numerous papers explore the TM empirically, many of its properties have not yet been analyzed mathematically. In this article, we analyze the convergence of the TM when input is non-linearly related to output by the XOR-operator. Our analysis reveals that the TM, with just two conjunctive clauses, can converge almost surely to reproducing XOR, learning from training data over an infinite time horizon. Furthermore, the analysis shows how the hyper-parameter T guides clause construction so that the clauses capture the distinct sub-patterns in the data. Our analysis of convergence for XOR thus lays the foundation for analyzing other more complex logical expressions. These analyses altogether, from a mathematical perspective, provide new insights on why TMs have obtained the state-of-the-art performance on several pattern recognition problems.

Inference at-the-edge using embedded machine learning models is associated with challenging trade-offs between resource metrics, such as energy and memory footprint, and the performance metrics, such as computation time and accuracy. In this work, we go beyond the conventional Neural Network based approaches to explore Tsetlin Machine (TM), an emerging machine learning algorithm, that uses learning automata to create propositional logic for classification. We use algorithm-hardware co-design to propose a novel methodology for training and inference of TM. The methodology, called REDRESS, comprises independent TM training and inference techniques to reduce the memory footprint of the resulting automata to target low and ultra-low power applications. The array of Tsetlin Automata (TA) holds learned information in the binary form as bits: {0,1}, called excludes and includes, respectively. REDRESS proposes a lossless TA compression method, called the include-encoding, that stores only the information associated with includes to achieve over 99% compression. This is enabled by a novel computationally minimal training procedure, called the Tsetlin Automata Re-profiling, to improve the accuracy and increase the sparsity of TA to reduce the number of includes, hence, the memory footprint. Finally, REDRESS includes an inherently bit-parallel inference algorithm that operates on the optimally trained TA in the compressed domain, that does not require decompression during runtime, to obtain high speedups when compared with the state-of-the-art Binary Neural Network (BNN) models. In this work, we demonstrate that using REDRESS approach, TM outperforms BNN models on all design metrics for five benchmark datasets viz. MNIST, CIFAR2, KWS6, Fashion-MNIST and Kuzushiji-MNIST. When implemented on an STM32F746G-DISCO microcontroller, REDRESS obtained speedups and energy savings ranging 5-5700× compared with different BNN models.

The state-of-the-art natural language processing models have raised the bar for excellent performance on a variety of tasks in recent years. However, concerns are rising over their primitive sensitivity to distribution biases that reside in the training and testing data. This issue hugely impacts the performance of the models when exposed to out-of-distribution and counterfactual data. The root cause seems to be that many machine learning models are prone to learn the shortcuts, modelling simple correlations rather than more fundamental and general relationships. As a result, such text classifiers tend to perform poorly when a human makes minor modifications to the data, which raises questions regarding their robustness. In this paper, we employ a rule-based architecture called Tsetlin Machine (TM) that learns both simple and complex correlations by ANDing features and their negations. As such, it generates explainable AND-rules using negated and non-negated reasoning. Here, we explore how non-negated reasoning can be more prone to distribution biases than negated reasoning. We further leverage this finding by adapting the TM architecture to mainly perform negated reasoning using the specificity parameter s. As a result, the AND-rules becomes robust to spurious correlations and can also correctly predict counterfactual data. Our empirical investigation of the model's robustness uses the specificity s to control the degree of negated reasoning. Experiments on publicly available Counterfactually-Augmented Data demonstrate that the negated clauses are robust to spurious correlations and outperform Naive Bayes, SVM, and Bi-LSTM by up to 20 %, and ELMo by almost 6 % on counterfactual test data.

Recent advancements in natural language processing (NLP) have reshaped the industry, with powerful language models such as GPT-3 achieving superhuman performance on various tasks. However, the increasing complexity of such models turns them into “black boxes”, creating uncertainty about their internal operation and decision-making. Tsetlin Machine (TM) employs human-interpretable conjunctive clauses in propositional logic to solve complex pattern recognition problems and has demonstrated competitive performance in various NLP tasks. In this paper, we propose ConvTextTM, a novel convolutional TM architecture for text classification. While legacy TM solutions treat the whole text as a corpus-specific set-of-words (SOW), ConvTextTM breaks down the text into a sequence of text fragments. The convolution over the text fragments opens up for local position-aware analysis. Further, ConvTextTM eliminates the dependency on a corpus-specific vocabulary. Instead, it employs a generic SOW formed by the tokenization scheme of the Bidirectional Encoder Representations from Transformers (BERT). The convolution binds together the tokens, allowing ConvTextTM to address the out-of-vocabulary problem as well as spelling errors. We investigate the local explainability of our proposed method using clause-based features. Extensive experiments are conducted on seven datasets, to demonstrate that the accuracy of ConvTextTM is either superior or comparable to state-of-the-art baselines.

The proliferation of fake news, i.e., news intentionally spread for misinformation, poses a threat to individuals and society. Despite various fact-checking websites such as PolitiFact, robust detection techniques are required to deal with the increase in fake news. Several deep learning models show promising results for fake news classification, however, their black-box nature makes it difficult to explain their classification decisions and quality-assure the models. We here address this problem by proposing a novel interpretable fake news detection framework based on the recently introduced Tsetlin Machine (TM). In brief, we utilize the conjunctive clauses of the TM to capture lexical and semantic properties of both true and fake news text. Further, we use clause ensembles to calculate the credibility of fake news. For evaluation, we conduct experiments on two publicly available datasets, PolitiFact and GossipCop, and demonstrate that the TM framework significantly outperforms previously published baselines by at least 5% in terms of accuracy, with the added benefit of an interpretable logic-based representation. In addition, our approach provides a higher F1-score than BERT and XLNet, however, we obtain slightly lower accuracy. We finally present a case study on our model’s explainability, demonstrating how it decomposes into meaningful words and their negations.

Abstract Recent research in novelty detection focuses mainly on document-level classification, employing deep neural networks (DNN). However, the black-box nature of DNNs makes it difficult to extract an exact explanation of why a document is considered novel. In addition, dealing with novelty at the word level is crucial to provide a more fine-grained analysis than what is available at the document level. In this work, we propose a Tsetlin Machine (TM)-based architecture for scoring individual words according to their contribution to novelty. Our approach encodes a description of the novel documents using the linguistic patterns captured by TM clauses. We then adapt this description to measure how much a word contributes to making documents novel. Our experimental results demonstrate how our approach breaks down novelty into interpretable phrases, successfully measuring novelty.

Abstract A wide range of applications in marine ecology extensively uses underwater cameras. Still, to efficiently process the vast amount of data generated, we need to develop tools that can automatically detect and recognize species captured on film. Classifying fish species from videos and images in natural environments can be challenging because of noise and variation in illumination and the surrounding habitat. In this paper, we propose a two-step deep learning approach for the detection and classification of temperate fishes without pre-filtering. The first step is to detect each single fish in an image, independent of species and sex. For this purpose, we employ the You Only Look Once (YOLO) object detection technique. In the second step, we adopt a Convolutional Neural Network (CNN) with the Squeeze-and-Excitation (SE) architecture for classifying each fish in the image without pre-filtering. We apply transfer learning to overcome the limited training samples of temperate fishes and to improve the accuracy of the classification. This is done by training the object detection model with ImageNet and the fish classifier via a public dataset (Fish4Knowledge), whereupon both the object detection and classifier are updated with temperate fishes of interest. The weights obtained from pre-training are applied to post-training as a priori. Our solution achieves the state-of-the-art accuracy of 99.27% using the pre-training model. The accuracies using the post-training model are also high; 83.68% and 87.74% with and without image augmentation, respectively. This strongly indicates that the solution is viable with a more extensive dataset.

The Tsetlin Machine (TM) is a recent machine learning algorithm with several distinct properties, such as interpretability, simplicity, and hardware-friendliness. Although numerous empirical evaluations report on its performance, the mathematical analysis of its convergence is still open. In this article, we analyze the convergence of the TM with only one clause involved for classification. More specifically, we examine two basic logical operators, namely, the "IDENTITY"- and "NOT" operators. Our analysis reveals that the TM, with just one clause, can converge correctly to the intended logical operator, learning from training data over an infinite time horizon. Besides, it can capture arbitrarily rare patterns and select the most accurate one when two candidate patterns are incompatible, by configuring a granularity parameter. The analysis of the convergence of the two basic operators lays the foundation for analyzing other logical operators. These analyses altogether, from a mathematical perspective, provide new insights on why TMs have obtained state-of-the-art performance on several pattern recognition problems.

The integration of reconfigurable intelligent surface (RIS) into millimeter wave (mmWave) vehicular communications offers the possibility to unleash the potential of future proliferating vehicular applications. However, the high-mobility-induced rapidly varying channel state information (CSI) has been making it challenging to obtain the accurate instantaneous CSI (I-CSI) and to cope with the incurable high signaling overhead. The situation may become worse when the RIS with a large number of passive reflecting elements is deployed. To overcome this challenge, we investigate in this paper a robust transmission scheme for the time-varying RIS-aided mmWave vehicular communications, in which, specifically, a multi-antenna base station (BS) serves vehicle user equipments (VUEs) with the help of RIS at the mmWave frequency. The uplink average achievable rate is maximized relying only upon the imperfect knowledge of statistical CSI. Considering the time-varying characteristics, we first propose an effective transmission protocol by reasonably configuring the time-scale of CSI acquisition in order to significantly relax the frequency of channel information updates, which constitutes one of the most critical issues in RIS-aided vehicular communications. Then, the formulated resource allocation problem is discussed in the single- and multi-VUE case, respectively. To be specific, for the single-VUE case, a closed-form expression of the average rate is derived by extracting the statistical characteristics of mmWave channels, and an alternating optimization (AO)-based algorithm is proposed. For the multi-VUE case, we develop an efficient algorithm, called JAPMC, to circumvent the unavailability of the closed-form of the objective function and probabilistic constraint by constructing quadratic surrogates of that. Simulation results confirm the effectiveness and robustness of our proposed algorithms as compared to benchmark schemes.

In 6G and future networks, joint optimization of communication and computational resources lays the foundation for various delay-sensitive intelligent IoT services in the fog computing architecture. In this paper, we present a multi-device collaborative computing architecture in the cell association environment to accelerate the processing procedure of data generated by smart IoT devices. In this scenario, a two-tier task scheduling scheme and an uplink and downlink power allocation factor are jointly optimized to reduce the data processing delay and improve fairness among different users, which is in nature a hard problem due to a series of non-convex constraints. To make the problem tractable, the problem is transformed into a smooth non-convex problem with the introduction of auxiliary variables and then decoupled into two subproblems based on the data transmission and processing procedure. Thereafter, different methods such as Successive Convex Approximation (SCA) and Block Successive Upperbound Minimization (BSUM) are employed to reconstruct several upper-bound convex optimization subproblems. Besides, a fast 0–1 binary offloading scheme is proposed based on the original algorithm. Finally, the simulation results depict the effectiveness of the proposed algorithms in detail, and the scalability of the system is also examined.

This paper proposes human-interpretable learning of aspect-based sentiment analysis (ABSA), employing the recently introduced Tsetlin Machines (TMs). We attain interpretability by converting the intricate position-dependent textual semantics into binary form, mapping all the features into bag-of-words (BOWs). The binary form BOWs are encoded so that the information on the aspect and context words are nearly lossless for sentiment classification. We further adapt the BOWs as input to the TM, enabling learning of aspect-based sentiment patterns in propositional logic. To evaluate interpretability and accuracy, we conducted experiments on two widely used ABSA datasets of SemEval 2014: Restaurant 14 and Laptop 14. The experiments show how each relevant feature takes part in conjunctive clauses that contain the context information for the corresponding aspect word, demonstrating human-level interpretability. At the same time, the obtained accuracy is competitive with existing neural network models, reaching 78.02% on Restaurant 14 and 73.51% on Laptop 14.

Most supervised text classification approaches assume a closed world, counting on all classes being present in the data at training time. This assumption can lead to unpredictable behaviour during operation, whenever novel, previously unseen, classes appear. Although deep learning-based methods have recently been used for novelty detection, they are challenging to interpret due to their black-box nature. This paper addresses \emph{interpretable} open-world text classification, where the trained classifier must deal with novel classes during operation. To this end, we extend the recently introduced Tsetlin machine (TM) with a novelty scoring mechanism. The mechanism uses the conjunctive clauses of the TM to measure to what degree a text matches the classes covered by the training data. We demonstrate that the clauses provide a succinct interpretable description of known topics, and that our scoring mechanism makes it possible to discern novel topics from the known ones. Empirically, our TM-based approach outperforms seven other novelty detection schemes on three out of five datasets, and performs second and third best on the remaining, with the added benefit of an interpretable propositional logic-based representation.

Using logical clauses to represent patterns, Tsetlin Machine (TM) have recently obtained competitive performance in terms of accuracy, memory footprint, energy, and learning speed on several benchmarks. Each TM clause votes for or against a particular class, with classification resolved using a majority vote. While the evaluation of clauses is fast, being based on binary operators, the voting makes it necessary to synchronize the clause evaluation, impeding parallelization. In this paper, we propose a novel scheme for desynchronizing the evaluation of clauses, eliminating the voting bottleneck. In brief, every clause runs in its own thread for massive native parallelism. For each training example, we keep track of the class votes obtained from the clauses in local voting tallies. The local voting tallies allow us to detach the processing of each clause from the rest of the clauses, supporting decentralized learning. This means that the TM most of the time will operate on outdated voting tallies. We evaluated the proposed parallelization across diverse learning tasks and it turns out that our decentralized TM learning algorithm copes well with working on outdated data, resulting in no significant loss in learning accuracy. Furthermore, we show that the approach provides up to 50 times faster learning. Finally, learning time is almost constant for reasonable clause amounts (employing from 20 to 7,000 clauses on a Tesla V100 GPU). For sufficiently large clause numbers, computation time increases approximately proportionally. Our parallel and asynchronous architecture thus allows processing of more massive datasets and operating with more clauses for higher accuracy.

Aspect-based sentiment analysis (ABSA) aims at identifying fine-grained polarity of opinion associated with a given aspect word. Several existing articles demonstrated promising ABSA accuracy using positional embedding to show the relationship between an aspect word and its context. In most cases, the positional embedding depends on the distance between the aspect word and the remaining words in the context, known as the position index sequence. However, these techniques usually employ both complex preprocessing approaches with additional trainable positional embedding and complex architectures to obtain the state-of-the-art performance. In this paper, we simplify preprocessing by including polarity lexicon replacement and masking techniques that carry the information of the aspect word’s position and eliminate the positional embedding. We then adopt a novel and concise architecture using two Bidirectional GRU along with an attention layer to classify the aspect based on its context words. Experiment results show that the simplified preprocessing and the concise architecture significantly improve the accuracy of the publicly available ABSA datasets, obtaining 81.37%, 75.39%, 80.88%, and 89.30% in restaurant 14, laptop 14, restaurant 15, and restaurant 16 respectively.

An optimization problem of content placement in cooperative caching is formulated, with the aim of maximizing the sum mean opinion score (MOS) of mobile users. Firstly, as user mobility and content popularity have significant impacts on the user experience, a recurrent neural network (RNN) is invoked for user mobility prediction and content popularity prediction. More particularly, practical data collected from GPS-tracker app on smartphones is tackled to test the accuracy of user mobility prediction. Then, based on the predicted mobile users' positions and content popularity, a learning automata based Q-learning (LAQL) algorithm for cooperative caching is proposed, in which learning automata (LA) is invoked for Q-learning to obtain an optimal action selection in a random and stationary environment. It is proven that the LA based action selection scheme is capable of enabling every state to select the optimal action with arbitrary high probability if Q-learning is able to converge to the optimal Q value eventually. In the LAQL algorithm, a central processor acts as the intelligent agent, which allocate contents to BSs according to the reward or penalty from the feedback of the BSs and users, iteratively. To characterize the performance of the proposed LAQL algorithms, sum MOS of users is applied to define the reward function. Extensive simulation results reveal that: 1) the prediction error of RNNs based algorithm lessen with the increase of iterations and nodes; 2) the proposed LAQL achieves significant performance improvement against traditional Q-learning algorithm; and 3) the cooperative caching scheme is capable of outperforming non-cooperative caching and random caching of 3% and 4%, respectively.

Although the field of learning automata (LA) has made significant progress in the past four decades, the LA-based methods to tackle problems involving environments with a large number of actions is, in reality, relatively unresolved. The extension of the traditional LA to problems within this domain cannot be easily established when the number of actions is very large. This is because the dimensionality of the action probability vector is correspondingly large, and so, most components of the vector will soon have values that are smaller than the machine accuracy permits, implying that they will never be chosen. This paper presents a solution that extends the continuous pursuit paradigm to such large-actioned problem domains. The beauty of the solution is that it is hierarchical, where all the actions offered by the environment reside as leaves of the hierarchy. Furthermore, at every level, we merely require a two-action LA that automatically resolves the problem of dealing with arbitrarily small action probabilities. In addition, since all the LA invoke the pursuit paradigm, the best action at every level trickles up toward the root. Thus, by invoking the property of the “max” operator, in which the maximum of numerous maxima is the overall maximum, the hierarchy of LA converges to the optimal action. This paper describes the scheme and formally proves its E-optimal convergence. The results presented here can, rather trivially, be extended for the families of discretized and Bayesian pursuit LA too. This paper also reports extensive experimental results (including for environments having 128 and 256 actions) that demonstrate the power of the scheme and its computational advantages. As far as we know, there are no comparable pursuitbased results in the field of LA. In some cases, the hierarchical continuous pursuit automaton requires less than 18% of the number of iterations than the benchmark LR-Ischeme, which is, by all metrics, phenomenal.

This paper deals with the finite-time analysis (FTA) of learning automata (LA), which is a topic for which very little work has been reported in the literature. This is as opposed to the asymptotic steady-state analysis for which there are, probably, scores of papers. As clarified later, unarguably, the FTA of Markov chains, in general, and of LA, in particular, is far more complex than the asymptotic steady-state analysis. Such an FTA provides rigid bounds for the time required for the LA to attain to a given convergence accuracy. We concentrate on the FTA of the Discretized Pursuit Automaton (DPA), which is probably one of the fastest and most accurate reported LA. Although such an analysis was carried out many years ago, we record that the previous work is flawed. More specifically, in all brevity, the flaw lies in the wrongly "derived" monotonic behavior of the LA after a certain number of iterations. Rather, we claim that the property should be invoked is the submartingale property. This renders the proof to be much more involved and deep. In this paper, we rectify the flaw and reestablish the FTA based on such a submartingale phenomenon. More importantly, from the derived analysis, we are able to discover and clarify, for the first time, the underlying dilemma between the DPA's exploitation and exploration properties. We also nontrivially confirm the existence of the optimal learning rate, which yields a better comprehension of the DPA itself.

To bridge the gap between the resource‐constrained mobile devices and the resource‐demanding applications, mobile cloud computing (MCC) emerges for offloading complex tasks to a cloud server. Based on this concept, cloudlets, which move available resource to the vicinity of the mobile network, enhance further the system accessibility and performance. Moreover, to strengthen the network capacity in traffic intensive area, dense small cell network (DSCN) is proposed as one of the promising solutions. In this study, the operation of cloudlets and DSCN is collaboratively studied in order to further improve the system performance. On the one hand, users can select a cloudlet and dynamically adapt the connection according to the performance and the cost, which is referred to as a user‐essential dynamic cloudlet selection problem. On the other hand, a cloudlet needs to set the optimal selling price and the size of resource for the users, which is considered as a cloudlet resource allocation problem. To jointly address the problems of dynamic cloudlet selection and resource allocation, the authors propose a hierarchical evolutionary game to maximise the utilities. Simulation studies are carried out to demonstrate the effectiveness of the proposed algorithms, which, indeed, improve the entire system performance significantly.

Color transfer in image processing usually suffers from misleading color mapping and loss of details. This paper presents a novel directive local color transfer method based on dynamic look-up table (D-DLT) to solve these problems in two steps. First, a directive mapping between the source and the reference image is established based on the salient detection and the color clusters to obtain directive color transfer intention. Then, dynamic look-up tables are created according to the color clusters to preserve the details, which can suppress pseudo contours and avoid detail loss. Subjective and objective assessments are presented to verify the feasibility and the availability of the proposed approach. Experimental results demonstrate that our proposed method has better performance on natural color images than classical color transfer algorithms. Furthermore, the reference image can be extended to color blocks instead of images.

During 2017, digital audio broadcasting (DAB) replaces frequency modulation (FM) broadcasting and becomes the only technology for national terrestrial audio broadcasting services in Norway. As Norway is the first country that replaces FM completely with DAB, it is of great importance to measure the signal strength of such a technology in massive deployments and to tune a simulation model as a reference for future studies. Therefore, field measurements of received signal strength are carried out in a typical Norwegian area in this work. Based on the data obtained from the measurements, a simulator with a recent empirical propagation model, namely, ITU-R P.1546-5, has been calibrated. The findings, in short, suggest that different clutter codes and attenuation parameters have to be tuned according to the measurement results in order to increase the precision of the simulation model. By properly tuning the parameters in the model, the precision has increased 91.7% compared with the default configuration. In addition, the tuned model is validated through another field measurement in a different area and obtains an increased precision of 72.9%.

Maintaining link stability among randomly deployed network nodes is one of the key challenges for effective communication in mobile ad hoc networks (MANETs). Under uniform speed and random trajectory of mobile nodes, there must be a unified model to determine an adequate strategy that addresses the issue of link stability in MANETs. We present a novel dynamic link connectivity (DLC) strategy that maintains link stability through efficient link connectivity among the neighboring nodes using Hello messaging. We also perform stochastic analysis of the proposed strategy, which predicts the future link status among the neighboring nodes at different time steps of a Markov process. We find that the link stability is affected by the received signal strength, signal‐to‐noise ratio, transition rates between the connection and disconnection states, and probabilities of link connectivity and disconnectivity at steady state. Analytical and simulation results indicate efficacy of the proposed strategy in terms of reduced communication overhead, lower propagation delay, and better energy efficiency of the network. The results also demonstrate that the proposed strategy minimizes the average response time, increases the throughput, and reduces the packet loss ratio, thereby, maintaining efficient link stability among the neighboring nodes.

Stochastic geometry tool is an accurate and tractable approach to analyse the performance characteristics of heterogeneous networks (HetNets) with unplanned deployment of base stations (BSs). In this study, the authors analyse the downlink coverage and capacity by taking into account the separation between the macro BS (MBS) and small BSs (SBSs) in a HetNet. MBSs, SBSs, and users are spatially distributed as independent Poisson point processes. The whole space is divided into inner and outer subspaces around the MBSs and a typical user is assumed to be in outer subspace. Analysis on the typical user in the outer subspace incorporates the effect of distance of SBSs from MBS. They derive the expressions for the coverage probability of an outer subspace typical user from SBS and MBS. They also derive the expressions for rate coverage of an outer subspace typical user from SBS and MBS. These metrics are analysed with varying inner subspace radius to incorporate the effect of BSs separation. Simulation results show that SBSs away from MBS in the coverage region of MBS provide good performance to their associated users in terms of coverage probability and rate coverage.

In heterogeneous cellular networks (HCNets) smallcells are overlaid with macro-cells to handle heavy cellular data traffic in an efficient way. To achieve fast and reliable access to data with a better coverage it is necessary to offload users to the underutilized small-cells from the congested macro-cells. Sharing of the same licensed frequency spectrum by smallcells and macro-cells results in heavy cross-tier interference which significantly affects the downlink SINR. In this paper, we investigate and analyze a joint frequency-division duplex based cross-tier complementary spectrum sharing technique which is also regarded as reverse frequency allocation (RFA) scheme with load balancing. This scheme mitigates inter-cell interference (ICI) and better balances the load across the network. It also guarantees enhanced spectral efficiency and coverage probability, particularly for cell edge users (CEUs). Numerical results indicate that load balancing together with RFA is necessary to enhance the coverage of CEU in co-channel HCNets.

Channel aggregation (CA) techniques can offer flexible channel allocation and improve overall system performance in multi-channel cognitive radio networks (CRNs). Although many CA techniques have been proposed and studied, the impact of guard-band on CA for channel access has not been addressed in-depth. In this paper, we study the guard-band allocation mechanisms in discontinuous-orthogonal frequency division multiplexing (D-OFDM) based CRNs, and investigate the impact of guard-band sharing on SU flows when CA is enabled. Continuous time Markov chain (CTMC) based models have been developed in order to investigate the stochastic behavior of PU and SU flows. Based on our mathematical analysis and simulation results, we observe that when guard-band is considered together with CA, the performance of the network varies according to different access schemes with distinct guard-band allocation mechanisms.

A precise prediction of domestic demand is very important for establishing home energy management system and preventing the damage caused by overloading. In this work, active and reactive power consumption prediction model based on historical power usage data and external environment parameter data (temperature and solar radiation) is presented for a typical Southern Norwegian house. In the presented model, a neural network is adopted as a main prediction technique and historical domestic load data of around 2 years are utilized for training and testing purpose. Temperature and global irradiation (which illustrates the solar radiation level quantitatively) are employed as external parameters. From the results, the efficiency of predictions are evaluated and compared. It can be observed from the numerical results that predictions using historical power data together with external data perform better than the case where only power usage data are adopted.

With the implementation of channel assembling (CA) techniques, higher data rate can be achieved for secondary users in multi-channel cognitive radio networks. Recent studies which are based on loss systems show that maximal capacity can be achieved using dynamic CA strategies. However the channel allocation schemes suffer from high blocking and forced termination when primary users become active. In this paper, we propose to introduce queues for secondary users so that those flows that would otherwise be blocked or forcibly terminated could be buffered and possibly served later. More specifically, in a multi-channel network with heterogeneous traffic, two queues are separately allocated to real-time and elastic users and channel access opportunities are distributed between these two queues in a way that real-time services receive higher priority. Two queuing schemes are introduced based on the delay tolerance of interrupted elastic services. Furthermore, continuous time Markov chain models are developed to evaluate the performance of the proposed CA strategy with queues, and the correctness as well as the preciseness of the derived theoretical models are verified through extensive simulations. Numerical results demonstrate that the integration of queues can further increase the capacity of the secondary network and spectrum utilization while decreasing blocking probability and forced termination probability.

Flexible channel allocation may be applied to multi-channel cognitive radio networks (CRNs) through either channel assembling (CA) or channel fragmentation (CF). While CA allows one secondary user (SU) occupy multiple channels when primary users (PUs) are absent, CF provides finer granularity for channel occupancy by allocating a portion of one channel to an SU flow. In this paper, we investigate the impact of CF together with CA for SU flows by proposing a channel access strategy which activates both CF and CA and correspondingly evaluating its performance. In addition, we also consider a novel scenario where CA is enabled for PU flows. The performance evaluation is conducted based on continuous time Markov chain (CTMC) modeling and simulations. Through mathematical analyses and simulation results, we demonstrate that higher system capacity can be achieved indeed by jointly employing both CA and CF, in comparison with the CA-only strategies. However, this benefit is obtained only under certain conditions which are pointed out in this paper. Furthermore, the theoretical capacity upper bound for SU flows with both CF and CA enabled is derived when PU activities are relatively static compared with SU flows.

We consider the problem of a user operating within a Cognitive Radio Network (CRN) which involves N channels each associated with a Primary User (PU). The problem consists of allocating a channel which, at any given time instant is not being used by a PU, to a Secondary User (SU). Within our study, we assume that a SU is allowed to perform “channel switching”, i.e., to choose an alternate channel S times (where S +1 ≤ N) if the previous choice does not lead to a channel which is vacant. The paper first presents a formal probabilistic model for the problem itself, referred to as the Formal Secondary Channel Selection (FSCS) problem, and the characteristics of the FSCS are then analyzed. Thereafter, the paper proposes a fascinating solution to the FSCS problem by invoking the recently devised Bayesian Learning Automaton (BLA). The crucial advantage of the BLA is that unlike traditional Learning Automata (LA), it does not involve an action probability vector, but rather relies on “sampling” as per the a posteriori Bayesian estimates of the channel occupation probabilities. However, rather than utilize the BLA in the form that was earlier proposed, we shall extend it to the so-called Switchable Bayesian Learning Automaton (SBLA), which, indeed, attains the optimal solution in the overall composite action space. Apart from proposing the solution, the paper also contains detailed simulation results which demonstrate the power of the solution proposed.

In this paper, we propose a novel routing protocol which is integrated with channel assignment for multi-channel multi-hop wireless ad hoc networks. In such a network, each node is equipped with three transceivers. One is always tuned on a control channel which is responsible for control and broadcast messages, and the other two perform as transmitter and receiver respectively for traffic flows on different data channels. The routing protocol works in an on-demand manner, and the proposed routing discovery process selects a path that potentially traverses nodes with lighter traffic load and lower number of carried flows. With a given number of non-overlapping channels, the optimal solution of channel allocation for multiple flows is shown to be in general NP-hard. Therefore, a heuristic algorithm for channel allocation based on the information acquired along a flow path is adopted, in order to mitigate two types of interference, i.e., inter-path interference and intra-path interference. NS2 based simulation experiments are conducted to evaluate the performance of the proposed protocol.

Cognitive radio has been introduced to increase spectrum utilization efficiency. To further improve bandwidth utilization of cognitive radio users, channel aggregation (CA) techniques can be adopted for spectrum access. In this paper, we analyze the complexity of three CA strategies, in terms of required amount of handshakes for channel adaptation due to primary and secondary user activities. Continuous time Markov chain models are developed to evaluate the total number of handshakes required per unit time by different CA strategies and the analytical results are validated by simulations. Numerical results reveal that the complexity of CA strategies depends on the capability and the design principle of spectrum adaptation.

In cognitive radio networks, channel aggregation techniques which combine several channels together as one channel have been proposed in many MAC protocols. In this paper, spectrum adaptation is proposed in channel aggregation and two strategies which dynamically adjust channel occupancy of ongoing traffic flows are further developed. The performance of these strategies is evaluated using continuous time Markov chain models. Moreover, models in the quasi-stationary regime are analyzed and the closed-form capacity expression is derived in this regime. Numerical results demonstrate that the capacity of the secondary network can be improved by using channel aggregation with spectrum adaptation.

Channel hopping based parallel rendezvous multichannel MAC protocols have several advantages since they do not need a control channel, require only one transceiver, and produce higher system capacity. However, channel hopping sequences in existing parallel rendezvous MAC protocols have been designed as irrelevant to channel data rates, leading to under-utilization of channel resources in multi-rate multi-channel networks. Considering that data rates among channels may be different, we propose a dynamic parallel rendezvous multichannel MAC protocol for synchronized cognitive radio networks in which the secondary users adjust their own distinct hopping sequences according to the datarates of the available channels, in a datarate-aware manner. A Markov chain based model has been developed to analyze the aggregate datarate of the proposed protocol. Numerical results show that the proposed method can improve the aggregate datarate significantly, compared with that of the existing parallel rendezvous MAC protocol.

This paper presents two MAC strategies for multi-hop cognitive radio networks in single radio multi-channel cases. Both strategies use one of the idle multiple channels for communication among secondary users, and the network will leave the current channel and jump to another channel as a group if any primary user appears. The first strategy is based on a pre-defined pattern that will always tune to the next available channel when primary user emerges. The second one is based on the concept of connected dominating set in which a backbone is formed in the network in order to keep the continuity of the communication. The strategies are evaluated in both homogeneous and heterogeneous channels by simulations. Numerical results show that higher throughput has been achieved with the first strategy in homogeneous channels but in heterogeneous channels the latter strategy performs much better.

Non line-of-sight (NLOS) propagation in range measurement is a key problem for mobile terminal localization. This paper proposes a low computational residual test (LCRT) algorithm that can identify the number of line-of-sight (LOS) transmissions and reduce the computational complexity compared with the residual test (RT) algorithm. LCRT is based on the assumption that when all range measurements are from LOS propagations, the normalized residual distribution follows the central chi-square distribution while for NLOS cases it is non-central. An optimized procedure to generate the sets of range measurements is adopted and least square (LS) instead of approximate maximum likelihood (AML) is used during the identification of LOS propagations, resulting in reduced computation complexity. Simulation results show that the LCRT can efficiently identify the set of LOS. The correct decision rate is higher than 92% and the variances of results are approaching to the Cramer-Rao lower bound (CRLB) when there are more than 3 LOS propagations.