Volume 11 (2021): Issue 1 (January 2021)

Music has the ability to evoke different emotions in people, which is reflected in their physiological signals. Advances in affective computing have introduced computational methods to analyse these signals and understand the relationship between music and emotion in greater detail. We analyse Electrodermal Activity (EDA), Blood Volume Pulse (BVP), Skin Temperature (ST) and Pupil Dilation (PD) collected from 24 participants while they listen to 12 pieces from 3 different genres of music. A set of 34 features were extracted from each signal and 6 different feature selection methods were applied to identify useful features. Empirical analysis shows that a neural network (NN) with a set of features extracted from the physiological signals can achieve 99.2% accuracy in differentiating among the 3 music genres. The model also reaches 98.5% accuracy in classification based on participants’ subjective rating of emotion. The paper also identifies some useful features to improve accuracy of the classification models. Furthermore, we introduce a new technique called ’Gingerbread Animation’ to visualise the physiological signals we record as a video, and to make these signals more comprehensible to the human eye, and also appropriate for computer vision techniques such as Convolutional Neural Networks (CNNs). Our results overall provide a strong motivation to investigate the relationship between physiological signals and music, which can lead to improvements in music therapy for mental health care and musicogenic epilepsy reduction (our long term goal).

The main purpose of a topological index is to encode a chemical structure by a number. A topological index is a graph invariant, which decribes the topology of the graph and remains constant under a graph automorphism. Topological indices play a wide role in the study of QSAR (quantitative structure-activity relationship) and QSPR (quantitative structure-property relationship). Topological indices are implemented to judge the bioactivity of chemical compounds. In this article, we compute the ABC (atom-bond connectivity); ABC₄ (fourth version of ABC), GA (geometric arithmetic) and GA₅ (fifth version of GA) indices of some networks sheet. These networks include: octonano window sheet; equilateral triangular tetra sheet; rectangular sheet; and rectangular tetra sheet networks.

Recurrent neural networks (RNN) have been successfully applied to various sequential decision-making tasks, natural language processing applications, and time-series predictions. Such networks are usually trained through back-propagation through time (BPTT) which is prohibitively expensive, especially when the length of the time dependencies and the number of hidden neurons increase. To reduce the training time, extreme learning machines (ELMs) have been recently applied to RNN training, reaching a 99% speedup on some applications. Due to its non-iterative nature, ELM training, when parallelized, has the potential to reach higher speedups than BPTT.

In this work, we present Opt-PR-ELM, an optimized parallel RNN training algorithm based on ELM that takes advantage of the GPU shared memory and of parallel QR factorization algorithms to efficiently reach optimal solutions. The theoretical analysis of the proposed algorithm is presented on six RNN architectures, including LSTM and GRU, and its performance is empirically tested on ten time-series prediction applications. Opt-PR-ELM is shown to reach up to 461 times speedup over its sequential counterpart and to require up to 20x less time to train than parallel BPTT. Such high speedups over new generation CPUs are extremely crucial in real-time applications and IoT environments.

The paper proposes summarized attribution-based post-hoc explanations for the detection and identification of bias in data. A global explanation is proposed, and a step-by-step framework on how to detect and test bias is introduced. Since removing unwanted bias is often a complicated and tremendous task, it is automatically inserted, instead. Then, the bias is evaluated with the proposed counterfactual approach. The obtained results are validated on a sample skin lesion dataset. Using the proposed method, a number of possible bias-causing artifacts are successfully identified and confirmed in dermoscopy images. In particular, it is confirmed that black frames have a strong influence on Convolutional Neural Network’s prediction: 22% of them changed the prediction from benign to malignant.

In this paper we propose a novel method for invariant image reconstruction with the properly selected degree of symmetry. We make use of Zernike radial moments to represent an image due to their invariance properties to isometry transformations and the ability to uniquely represent the salient features of the image. The regularized ridge regression estimation strategy under symmetry constraints for estimating Zernike moments is proposed. This extended regularization problem allows us to enforces the bilateral symmetry in the reconstructed object. This is achieved by the proper choice of two regularization parameters controlling the level of reconstruction accuracy and the acceptable degree of symmetry. As a byproduct of our studies we propose an algorithm for estimating an angle of the symmetry axis which in turn is used to determine the possible asymmetry present in the image. The proposed image recovery under the symmetry constraints model is tested in a number of experiments involving image reconstruction and symmetry estimation.