This is Part II of the two-part comprehensive survey devoted to a computing framework most commonly known under the names Hyperdimensional Computing and Vector Symbolic Architectures (HDC/VSA). Both names refer to a family of computational models that use high-dimensional distributed representations and rely on the algebraic properties of their key operations to incorporate the advantages of structured symbolic representations and vector distributed representations. Holographic Reduced Representations [321, 326] is an influential HDC/VSA model that is well known in the machine learning domain and often used to refer to the whole family. However, for the sake of consistency, we use HDC/VSA to refer to the field.Part I of this survey [222] covered foundational aspects of the field, such as the historical context leading to the development of HDC/VSA, key elements of any HDC/VSA model, known HDC/VSA models, and the transformation of input data of various types into high-dimensional vectors suitable for HDC/VSA. This second part surveys existing applications, the role of HDC/VSA in cognitive computing and architectures, as well as directions for future work. Most of the applications lie within the Machine Learning/Artificial Intelligence domain; however, we also cover other applications to provide a complete picture. The survey is written to be useful for both newcomers and practitioners.

The deployment of machine learning algorithms on resource-constrained edge devices is an important challenge from both theoretical and applied points of view. In this brief, we focus on resource-efficient randomly connected neural networks known as random vector functional link (RVFL) networks since their simple design and extremely fast training time make them very attractive for solving many applied classification tasks. We propose to represent input features via the density-based encoding known in the area of stochastic computing and use the operations of binding and bundling from the area of hyperdimensional computing for obtaining the activations of the hidden neurons. Using a collection of 121 real-world data sets from the UCI machine learning repository, we empirically show that the proposed approach demonstrates higher average accuracy than the conventional RVFL. We also demonstrate that it is possible to represent the readout matrix using only integers in a limited range with minimal loss in the accuracy. In this case, the proposed approach operates only on small n-bits integers, which results in a computationally efficient architecture. Finally, through hardware field-programmable gate array (FPGA) implementations, we show that such an approach consumes approximately 11 times less energy than that of the conventional RVFL.

Objective: The 2017 PhysioNet/CinC Challenge focused on automatic classification of atrial fibrillation (AF) in short ECGs. This study aimed to evaluate the use of the data and results from the challenge for detection of AF in longer ECGs, taken from three other PhysioNet datasets. Approach: The used data-driven models were based on features extracted from ECG recordings, calculated according to three solutions from the challenge. A Random Forest classifier was trained with the data from the challenge. The performance was evaluated on all non-overlapping 30 s segments in all recordings from three MIT-BIH datasets. Fifty-six models were trained using different feature sets, both before and after applying three feature reduction techniques. Main Results: Based on rhythm annotations, the AF proportion was 0.00 in the MIT-BIH Normal Sinus Rhythm (N = 46083 segments), 0.10 in the MIT-BIH Arrhythmia (N = 2880), and 0.41 in the MIT-BIH Atrial Fibrillation (N = 28104) dataset. For the best performing model, the corresponding detected proportions of AF were 0.00, 0.11 and 0.36 using all features, and 0.01, 0.10 and 0.38 when using the 15 best performing features. Significance: The results obtained on the MIT-BIH datasets indicate that the training data and solutions from the 2017 Physionet/Cinc Challenge can be useful tools for developing robust AF detectors also in longer ECG recordings, even when using a low number of carefully selected features. The use of feature selection allows significantly reducing the number of features while preserving the classification performance, which can be important when building low-complexity AF classifiers on ECG devices with constrained computational and energy resources.

This paper presents the use of automated machine learning for solving a practical problem of a real-life Smart Waste Management system. In particular, the focus of the paper is on the problem of detection (i.e., binary classification) of emptying of a recycling container using sensor measurements. Numerous data-driven methods for solving the problem are investigated in a realistic setting where most of the events are not actual emptying. The investigated methods include the existing manually engineered model and its modification as well as conventional machines learning algorithms. The use of machine learning allows improving the classification accuracy and recall of the existing manually engineered model from $86.8\%$ and $47.9\%$ to $99.1\%$ and $98.2\%$ , respectively, when using the best performing solution. This solution uses a Random Forest classifier on a set of features based on the filling level at different given time spans. Finally, compared to the baseline existing manually engineered model, the best performing solution also improves the quality of forecasts for emptying time of recycling containers.

A Bloom filter is a special case of an artificial neural network with two layers. Traditionally, it is seen as a simple data structure supporting membership queries on a set. The standard Bloom filter does not support the delete operation, and therefore, many applications use a counting Bloom filter to enable deletion. This paper proposes a generalization of the counting Bloom filter approach, called “autoscaling Bloom filters”, which allows adjustment of its capacity with probabilistic bounds on false positives and true positives. Thus, by relaxing the requirement on perfect true positive rate, the proposed autoscaling Bloom filter addresses the major difficulty of Bloom filters with respect to their scalability. In essence, the autoscaling Bloom filter is a binarized counting Bloom filter with an adjustable binarization threshold. We present the mathematical analysis of its performance and provide a procedure for minimizing its false positive rate.

Objective: Autonomic function during deep breathing (DB) is normally scored based on the assumption that the heart rate is synchronized with the breathing. We have observed individuals with subtle arrhythmias during DB where autonomic function cannot be evaluated. This study presents a novel method for analyzing cardiorespiratory synchronization: feature-based analysis of the similarity between heart rate and respiration using principles of hyperdimensional computing. Methods: Heart rate and respiration signals were modeled using Fourier series analysis. Three feature variables were derived and mapped to binary vectors in a high-dimensional space. Using both synthesized data and recordings from patients/healthy subjects, the similarity between the feature vectors was assessed using Hamming distance (high-dimensional space), Euclidean distance (original space), and with a coherence-based index. Methods were evaluated via classification of the similarity indices into three groups. Results: The distance-based methods achieved good separation of signals into classes with different degree of cardiorespiratory synchronization, also providing identification of patients with low cardiorespiratory synchronization but high values of conventional DB scores. Moreover, binary high-dimensional vectors allowed an additional analysis of the obtained Hamming distance. Conclusions: Feature-based similarity analysis using hyperdimensional computing is capable of identifying signals with low cardiorespiratory synchronization during DB due to arrhythmias. Vector-based similarity analysis could be applied to other types of feature variables than based on spectral analysis. Significance: The proposed methods for robustly assessing cardiorespiratory synchronization during DB facilitate the identification of individuals where the evaluation of autonomic function is problematic or even impossible, thus, increasing the correctness of the conventional DB scores.

This paper presents an approach for substantial reduction of the training and operating phases of Self-Organizing Maps in tasks of 2-D projection of multi-dimensional symbolic data for natural language processing such as language classification, topic extraction, and ontology development. The conventional approach for this type of problem is to use n-gram statistics as a fixed size representation for input of Self-Organizing Maps. The performance bottleneck with n-gram statistics is that the size of representation and as a result the computation time of Self-Organizing Maps grows exponentially with the size of n-grams. The presented approach is based on distributed representations of structured data using principles of hyperdimensional computing. The experiments performed on the European languages recognition task demonstrate that Self-Organizing Maps trained with distributed representations require less computations than the conventional n-gram statistics while well preserving the overall performance of Self-Organizing Maps. 

The Self-Organizing Map algorithm has been proven and demonstrated to be a useful paradigm for unsupervised machine learning of two-dimensional projections of multidimensional data. The tri-state Self-Organizing Maps have been proposed as an accelerated resource-efficient alternative to the Self-Organizing Maps for implementation on field-programmable gate array (FPGA) hardware. This paper presents a generalization of the tri-state Self-Organizing Maps. The proposed generalization, which we call integer Self-Organizing Maps, requires only integer operations for weight updates. The presented experiments demonstrated that the integer Self-Organizing Maps achieve better accuracy in a classification task when compared to the original tri-state Self-Organizing Maps.

Field-Programmable Gate Arrays (FPGA) are hardware components that hold several desirable properties for wearable and Internet of Things (IoT) devices. They offer hardware implementations of algorithms using parallel computing, which can be used to increase battery life or achieve short response-times. Further, they are re-programmable and can be made small, power-efficient and inexpensive. In this paper we propose a classifier targeted specifically for implementation on FPGAs by using principles from hyperdimensional computing and cellular automata. The proposed algorithm is shown to perform on par with Naive Bayes for two benchmark datasets while also being robust to noise. It is also synthesized to a commercially available off-the-shelf FPGA reaching over 57.1 million classifications per second for a 3-class problem using 40 input features of 8 bits each. The results in this paper show that the proposed classifier could be a viable option for applications demanding low power-consumption, fast real-time responses, or a robustness against post-training noise.

We propose a method for calculating the parameters of the resulting discrete channel for secondary users in cognitive radio systems formed by the algorithm for minimizing channel switchings. The availability of channel slots is defined by a simple Markov chain. We obtain mathematical expressions for determining the transition probabilities of a graph reduced to two states for any number of primary channels.