Coverage-Based Designs Improve Sample Mining and Hyperparameter Optimization.

Sampling one or more effective solutions from large search spaces is a recurring idea in machine learning (ML), and sequential optimization has become a popular solution. Typical examples include data summarization, sample mining for predictive modeling, and hyperparameter optimization. Existing solutions attempt to adaptively trade off between global exploration and local exploitation, in which the initial exploratory sample is critical to their success. While discrepancy-based samples have become the de facto approach for exploration, results from computer graphics suggest that coverage-based designs, e.g., Poisson disk sampling, can be a superior alternative. In order to successfully adopt coverage-based sample designs to ML applications, which were originally developed for 2-D image analysis, we propose fundamental advances by constructing a parameterized family of designs with provably improved coverage characteristics and developing algorithms for effective sample synthesis. Using experiments in sample mining and hyperparameter optimization for supervised learning, we show that our approach consistently outperforms the existing exploratory sampling methods in both blind exploration and sequential search with Bayesian optimization.

DDxNet: a deep learning model for automatic interpretation of electronic health records, electrocardiograms and electroencephalograms.

Effective patient care mandates rapid, yet accurate, diagnosis. With the abundance of non-invasive diagnostic measurements and electronic health records (EHR), manual interpretation for differential diagnosis has become time-consuming and challenging. This has led to wide-spread adoption of AI-powered tools, in pursuit of improving accuracy and efficiency of this process. While the unique challenges presented by each modality and clinical task demand customized tools, the cumbersome process of making problem-specific choices has triggered the critical need for a generic solution to enable rapid development of models in practice. In this spirit, we develop DDxNet, a deep architecture for time-varying clinical data, which we demonstrate to be well-suited for diagnostic tasks involving different modalities (ECG/EEG/EHR), required level of characterization (abnormality detection/phenotyping) and data fidelity (single-lead ECG/22-channel EEG). Using multiple benchmark problems, we show that DDxNet produces high-fidelity predictive models, and sometimes even provides significant performance gains over problem-specific solutions.

GrAMME: Semisupervised Learning Using Multilayered Graph Attention Models.

Modern data analysis pipelines are becoming increasingly complex due to the presence of multiview information sources. While graphs are effective in modeling complex relationships, in many scenarios, a single graph is rarely sufficient to succinctly represent all interactions, and hence, multilayered graphs have become popular. Though this leads to richer representations, extending solutions from the single-graph case is not straightforward. Consequently, there is a strong need for novel solutions to solve classical problems, such as node classification, in the multilayered case. In this article, we consider the problem of semisupervised learning with multilayered graphs. Though deep network embeddings, e.g., DeepWalk, are widely adopted for community discovery, we argue that feature learning with random node attributes, using graph neural networks, can be more effective. To this end, we propose to use attention models for effective feature learning and develop two novel architectures, GrAMME-SG and GrAMME-Fusion, that exploit the interlayer dependences for building multilayered graph embeddings. Using empirical studies on several benchmark data sets, we evaluate the proposed approaches and demonstrate significant performance improvements in comparison with the state-of-the-art network embedding strategies. The results also show that using simple random features is an effective choice, even in cases where explicit node attributes are not available.

Articulation constrained learning with application to speech emotion recognition.

Speech emotion recognition methods combining articulatory information with acoustic features have been previously shown to improve recognition performance. Collection of articulatory data on a large scale may not be feasible in many scenarios, thus restricting the scope and applicability of such methods. In this paper, a discriminative learning method for emotion recognition using both articulatory and acoustic information is proposed. A traditional ℓ 1-regularized logistic regression cost function is extended to include additional constraints that enforce the model to reconstruct articulatory data. This leads to sparse and interpretable representations jointly optimized for both tasks simultaneously. Furthermore, the model only requires articulatory features during training; only speech features are required for inference on out-of-sample data. Experiments are conducted to evaluate emotion recognition performance over vowels /AA/,/AE/,/IY/,/UW/ and complete utterances. Incorporating articulatory information is shown to significantly improve the performance for valence-based classification. Results obtained for within-corpus and cross-corpus categorical emotion recognition indicate that the proposed method is more effective at distinguishing happiness from other emotions.

Optimizing Kernel Machines Using Deep Learning.

Building highly nonlinear and nonparametric models is central to several state-of-the-art machine learning systems. Kernel methods form an important class of techniques that induce a reproducing kernel Hilbert space (RKHS) for inferring non-linear models through the construction of similarity functions from data. These methods are particularly preferred in cases where the training data sizes are limited and when prior knowledge of the data similarities is available. Despite their usefulness, they are limited by the computational complexity and their inability to support end-to-end learning with a task-specific objective. On the other hand, deep neural networks have become the de facto solution for end-to-end inference in several learning paradigms. In this paper, we explore the idea of using deep architectures to perform kernel machine optimization, for both computational efficiency and end-to-end inferencing. To this end, we develop the deep kernel machine optimization framework, that creates an ensemble of dense embeddings using Nyström kernel approximations and utilizes deep learning to generate task-specific representations through the fusion of the embeddings. Intuitively, the filters of the network are trained to fuse information from an ensemble of linear subspaces in the RKHS. Furthermore, we introduce the kernel dropout regularization to enable improved training convergence. Finally, we extend this framework to the multiple kernel case, by coupling a global fusion layer with pretrained deep kernel machines for each of the constituent kernels. Using case studies with limited training data, and lack of explicit feature sources, we demonstrate the effectiveness of our framework over conventional model inferencing techniques.

Empirically Estimable Classification Bounds Based on a Nonparametric Divergence Measure.

Information divergence functions play a critical role in statistics and information theory. In this paper we show that a non-parametric f-divergence measure can be used to provide improved bounds on the minimum binary classification probability of error for the case when the training and test data are drawn from the same distribution and for the case where there exists some mismatch between training and test distributions. We confirm the theoretical results by designing feature selection algorithms using the criteria from these bounds and by evaluating the algorithms on a series of pathological speech classification tasks.

Learning Stable Multilevel Dictionaries for Sparse Representations.

Sparse representations using learned dictionaries are being increasingly used with success in several data processing and machine learning applications. The increasing need for learning sparse models in large-scale applications motivates the development of efficient, robust, and provably good dictionary learning algorithms. Algorithmic stability and generalizability are desirable characteristics for dictionary learning algorithms that aim to build global dictionaries, which can efficiently model any test data similar to the training samples. In this paper, we propose an algorithm to learn dictionaries for sparse representations from large scale data, and prove that the proposed learning algorithm is stable and generalizable asymptotically. The algorithm employs a 1-D subspace clustering procedure, the K-hyperline clustering, to learn a hierarchical dictionary with multiple levels. We also propose an information-theoretic scheme to estimate the number of atoms needed in each level of learning and develop an ensemble approach to learn robust dictionaries. Using the proposed dictionaries, the sparse code for novel test data can be computed using a low-complexity pursuit procedure. We demonstrate the stability and generalization characteristics of the proposed algorithm using simulations. We also evaluate the utility of the multilevel dictionaries in compressed recovery and subspace learning applications.

Modeling Pathological Speech Perception From Data With Similarity Labels.

The current state of the art in judging pathological speech intelligibility is subjective assessment performed by trained speech pathologists (SLP). These tests, however, are inconsistent, costly and, oftentimes suffer from poor intra- and inter-judge reliability. As such, consistent, reliable, and perceptually-relevant objective evaluations of pathological speech are critical. Here, we propose a data-driven approach to this problem. We propose new cost functions for examining data from a series of experiments, whereby we ask certified SLPs to rate pathological speech along the perceptual dimensions that contribute to decreased intelligibility. We consider qualitative feedback from SLPs in the form of comparisons similar to statements "Is Speaker A's rhythm more similar to Speaker B or Speaker C?" Data of this form is common in behavioral research, but is different from the traditional data structures expected in supervised (data matrix + class labels) or unsupervised (data matrix) machine learning. The proposed method identifies relevant acoustic features that correlate with the ordinal data collected during the experiment. Using these features, we show that we are able to develop objective measures of the speech signal degradation that correlate well with SLP responses.

Multiple kernel sparse representations for supervised and unsupervised learning.

In complex visual recognition tasks, it is typical to adopt multiple descriptors, which describe different aspects of the images, for obtaining an improved recognition performance. Descriptors that have diverse forms can be fused into a unified feature space in a principled manner using kernel methods. Sparse models that generalize well to the test data can be learned in the unified kernel space, and appropriate constraints can be incorporated for application in supervised and unsupervised learning. In this paper, we propose to perform sparse coding and dictionary learning in the multiple kernel space, where the weights of the ensemble kernel are tuned based on graph-embedding principles such that class discrimination is maximized. In our proposed algorithm, dictionaries are inferred using multiple levels of 1D subspace clustering in the kernel space, and the sparse codes are obtained using a simple levelwise pursuit scheme. Empirical results for object recognition and image clustering show that our algorithm outperforms existing sparse coding based approaches, and compares favorably to other state-of-the-art methods.

Characterizing the distribution of the quadrilateral vowel space area.

The vowel space area (VSA) has been studied as a quantitative index of intelligibility to the extent it captures articulatory working space and reductions therein. The majority of such studies have been empirical wherein measures of VSA are correlated with perceptual measures of intelligibility. However, the literature contains minimal mathematical analysis of the properties of this metric. This paper further develops the theoretical underpinnings of this metric by presenting a detailed analysis of the statistical properties of the VSA and characterizing its distribution through the moment generating function. The theoretical analysis is confirmed by a series of experiments where empirically estimated and theoretically predicted statistics of this function are compared. The results show that on the Hillenbrand and TIMIT data, the theoretically predicted values of the higher-order statistics of the VSA match very well with the empirical estimates of the same.

Automatic assessment of vowel space area.

Vowel space area (VSA) is an attractive metric for the study of speech production deficits and reductions in intelligibility, in addition to the traditional study of vowel distinctiveness. Traditional VSA estimates are not currently sufficiently sensitive to map to production deficits. The present report describes an automated algorithm using healthy, connected speech rather than single syllables and estimates the entire vowel working space rather than corner vowels. Analyses reveal a strong correlation between the traditional VSA and automated estimates. When the two methods diverge, the automated method seems to provide a more accurate area since it accounts for all vowels.

Derivation of a new banded waveguide model topology for sound synthesis.

Banded waveguide (BWG) synthesis is an efficient method for real-time physical modeling of dispersive and multidimensional sounding objects, affording simulation of complex interactions, such as bowing. Current implementations, however, use nonphysical design parameters and produce a range of outputs that do not match equivalently designed modal and digital waveguide (DWG) models. This letter proposes a new topology for implementing BWG models without arbitrary parameters. The impulse response of the proposed model is identical to that of equivalent Karplus-Strong type and lumped modal models. Test of a nonlinear bi-directional bowed-string model demonstrates improved attack characteristics relative to prior BWG models.

Selecting Disorder-Specific Features for Speech Pathology Fingerprinting.

The general aim of this work is to learn a unique statistical signature for the state of a particular speech pathology. We pose this as a speaker identification problem for dysarthric individuals. To that end, we propose a novel algorithm for feature selection that aims to minimize the effects of speaker-specific features (e.g., fundamental frequency) and maximize the effects of pathology-specific features (e.g., vocal tract distortions and speech rhythm). We derive a cost function for optimizing feature selection that simultaneously trades off between these two competing criteria. Furthermore, we develop an efficient algorithm that optimizes this cost function and test the algorithm on a set of 34 dysarthric and 13 healthy speakers. Results show that the proposed method yields a set of features related to the speech disorder and not an individual's speaking style. When compared to other feature-selection algorithms, the proposed approach results in an improvement in a disorder fingerprinting task by selecting features that are specific to the disorder.

Porphyrins-Functionalized Single-Walled Carbon Nanotubes Chemiresistive Sensor Arrays for VOCs.

Single-walled carbon nanotubes (SWNTs) have been used extensively for sensor fabrication due to its high surface to volume ratio, nanosized structure and interesting electronic property. Lack of selectivity is a major limitation for SWNTs-based sensors. However, surface modification of SWNTs with a suitable molecular recognition system can enhance the sensitivity. On the other hand, porphyrins have been widely investigated as functional materials for chemical sensor fabrication due to their several unique and interesting physico-chemical properties. Structural differences between free-base and metal substituted porphyrins make them suitable for improving selectivity of sensors. However, their poor conductivity is an impediment in fabrication of prophyrin-based chemiresistor sensors. The present attempt is to resolve these issues by combining freebase- and metallo-porphyrins with SWNTs to fabricate SWNTs-porphyrin hybrid chemiresistor sensor arrays for monitoring volatile organic carbons (VOCs) in air. Differences in sensing performance were noticed for porphyrin with different functional group and with different central metal atom. The mechanistic study for acetone sensing was done using field-effect transistor (FET) measurements and revealed that the sensing mechanism of ruthenium octaethyl porphyrin hybrid device was governed by electrostatic gating effect, whereas iron tetraphenyl porphyrin hybrid device was governed by electrostatic gating and Schottky barrier modulation in combination. Further, the recorded electronic responses for all hybrid sensors were analyzed using a pattern-recognition analysis tool. The pattern-recognition analysis confirmed a definite pattern in response for different hybrid material and could efficiently differentiate analytes from one another. This discriminating capability of the hybrid nanosensor devices open up the possibilities for further development of highly dense nanosensor array with suitable porphyrin for E-nose application.

Information flow and application to epileptogenic focus localization from intracranial EEG.

Transfer entropy ( TE) is a recently proposed measure of the information flow between coupled linear or nonlinear systems. In this study, we suggest improvements in the selection of parameters for the estimation of TE that significantly enhance its accuracy and robustness in identifying the direction and the level of information flow between observed data series generated by coupled complex systems. We show the application of the improved TE method to long (in the order of days; approximately a total of 600 h across all patients), continuous, intracranial electroencephalograms (EEG) recorded in two different medical centers from four patients with focal temporal lobe epilepsy (TLE) for localization of their foci. All patients underwent ablative surgery of their clinically assessed foci. Based on a surrogate statistical analysis of the TE results, it is shown that the identified potential focal sites through the suggested analysis were in agreement with the clinically assessed sites of the epileptogenic focus in all patients analyzed. It is noteworthy that the analysis was conducted on the available whole-duration multielectrode EEG, that is, without any subjective prior selection of EEG segments or electrodes for analysis. The above, in conjunction with the use of surrogate data, make the results of this analysis robust. These findings suggest a critical role TE may play in epilepsy research in general, and as a tool for robust localization of the epileptogenic focus/foci in patients with focal epilepsy in particular.