Self-Supervised Open-Set Speaker Recognition with Laguerre-Voronoi Descriptors.

Speaker recognition is a challenging problem in behavioral biometrics that has been rigorously investigated over the last decade. Although numerous supervised closed-set systems inherit the power of deep neural networks, limited studies have been made on open-set speaker recognition. This paper proposes a self-supervised open-set speaker recognition that leverages the geometric properties of speaker distribution for accurate and robust speaker verification. The proposed framework consists of a deep neural network incorporating a wider viewpoint of temporal speech features and Laguerre-Voronoi diagram-based speech feature extraction. The deep neural network is trained with a specialized clustering criterion that only requires positive pairs during training. The experiments validated that the proposed system outperformed current state-of-the-art methods in open-set speaker recognition and cluster representation.

Machine learning prediction of atrial fibrillation in cardiovascular patients using cardiac magnetic resonance and electronic health information.

Background: Atrial fibrillation (AF) is a commonly encountered cardiac arrhythmia associated with morbidity and substantial healthcare costs. While patients with cardiovascular disease experience the greatest risk of new-onset AF, no risk model has been developed to predict AF occurrence in this population. We hypothesized that a patient-specific model could be delivered using cardiovascular magnetic resonance (CMR) disease phenotyping, contextual patient health information, and machine learning. Methods: Nine thousand four hundred forty-eight patients referred for CMR imaging were enrolled and followed over a 5-year period. Seven thousand, six hundred thirty-nine had no prior history of AF and were eligible to train and validate machine learning algorithms. Random survival forests (RSFs) were used to predict new-onset AF and compared to Cox proportional-hazard (CPH) models. The best performing features were identified from 115 variables sourced from three data domains: (i) CMR-based disease phenotype, (ii) patient health questionnaire, and (iii) electronic health records. We evaluated discriminative performance of optimized models using C-index and time-dependent AUC (tAUC). Results: A RSF-based model of 20 variables (CIROC-AF-20) delivered an overall C-index of 0.78 for the prediction of new-onset AF with respective tAUCs of 0.80, 0.79, and 0.78 at 1-, 2- and 3-years. This outperformed a novel CPH-based model and historic AF risk scores. At 1-year of follow-up, validation cohort patients classified as high-risk of future AF by CIROC-AF-20 went on to experience a 17.3% incidence of new-onset AF, being 24.7-fold higher risk than low risk patients. Conclusions: Using phenotypic data available at time of CMR imaging we developed and validated the first described risk model for the prediction of new-onset AF in patients with cardiovascular disease. Complementary value was provided by variables from patient-reported measures of health and the electronic health record, illustrating the value of multi-domain phenotypic data for the prediction of AF.

KinectGaitNet: Kinect-Based Gait Recognition Using Deep Convolutional Neural Network.

Over the past decade, gait recognition had gained a lot of attention in various research and industrial domains. These include remote surveillance, border control, medical rehabilitation, emotion detection from posture, fall detection, and sports training. The main advantages of identifying a person by their gait include unobtrusiveness, acceptance, and low costs. This paper proposes a convolutional neural network KinectGaitNet for Kinect-based gait recognition. The 3D coordinates of each of the body joints over the gait cycle are transformed to create a unique input representation. The proposed KinectGaitNet is trained directly using the 3D input representation without the necessity of the handcrafted features. The KinectGaitNet design allows avoiding gait cycle resampling, and the residual learning method ensures high accuracy without the degradation problem. The proposed deep learning architecture surpasses the recognition performance of all state-of-the-art methods for Kinect-based gait recognition by achieving 96.91% accuracy on UPCV and 99.33% accuracy on the KGB dataset. The method is the first, to the best of our knowledge, deep learning-based architecture that is based on a unique 3D input representation of joint coordinates. It achieves performance higher than previous traditional and deep learning methods, with fewer parameters and shorter inference time.

A multimethod approach to the differentiation of enthesis bone microstructure based on soft tissue type.

Whereas there is a wealth of research studying the nature of various soft tissues that attach to bone, comparatively little research focuses on the bone's microscopic properties in the area where these tissues attach. Using scanning electron microscopy to generate a dataset of 1600 images of soft tissue attachment sites, an image classification program with novel convolutional neural network architecture can categorize images of attachment areas by soft tissue type based on observed patterns in microstructure morphology. Using stained histological thin section and liquid crystal cross-polarized microscopy, it is determined that soft tissue type can be quantitatively determined from the microstructure. The primary diagnostic characters are the orientation of collagen fibers and heterogeneity of collagen density throughout the attachment area thickness. These determinations are made across broad taxonomic sampling and multiple skeletal elements.

Multimodal biometric system using rank-level fusion approach.

In many real-world applications, unimodal biometric systems often face significant limitations due to sensitivity to noise, intraclass variability, data quality, nonuniversality, and other factors. Attempting to improve the performance of individual matchers in such situations may not prove to be highly effective. Multibiometric systems seek to alleviate some of these problems by providing multiple pieces of evidence of the same identity. These systems help achieve an increase in performance that may not be possible using a single-biometric indicator. This paper presents an effective fusion scheme that combines information presented by multiple domain experts based on the rank-level fusion integration method. The developed multimodal biometric system possesses a number of unique qualities, starting from utilizing principal component analysis and Fisher's linear discriminant methods for individual matchers (face, ear, and signature) identity authentication and utilizing the novel rank-level fusion method in order to consolidate the results obtained from different biometric matchers. The ranks of individual matchers are combined using the highest rank, Borda count, and logistic regression approaches. The results indicate that fusion of individual modalities can improve the overall performance of the biometric system, even in the presence of low quality data. Insights on multibiometric design using rank-level fusion and its performance on a variety of biometric databases are discussed in the concluding section.