Choudur Lakshminarayan

Time-Based Compression and Classification of Heartbeats

Heart function measured by electrocardiograms (ECG) is crucial for patient care. ECG generated waveforms are used to find patterns of irregularities in cardiac cycles in patients. In many cases, irregularities evolve over an extended period of time that requires continuous monitoring. However, this requires wireless ECG recording devices. These devices consist of an enclosed system that includes electrodes, processing circuitry, and a wireless communication block imposing constraints on area, power, bandwidth, and resolution. In order to provide continuous monitoring of cardiac functions for real-time diagnostics, we propose a methodology that combines compression and analysis of heartbeats. The signal encoding scheme is the time-based integrate and fire sampler. The diagnostics can be performed directly on the samples avoiding reconstruction required by the competing finite rate of innovation and compressed sensing. As an added benefit, our scheme provides an efficient hardware implementation and a compressed representation for the ECG recordings, while still preserving discriminative features. We demonstrate the performance of our approach through a heartbeat classification application consisting of normal and irregular heartbeats known as arrhythmia. Our approach that uses simple features extracted from ECG signals is comparable to results in the published literature.

Nearest Neighbor Distributions for imbalanced classification

The class imbalance problem is pervasive in machine learning. To accurately classify the minority class, current methods rely on sampling schemes to close the gap between classes, or on the application of error costs to create algorithms which favor the minority class. Since the sampling schemes and costs must be specified, these methods are highly dependent on the class distributions present in the training set. This makes them difficult to apply in settings where the level of imbalance changes, such as in online streaming data. Often they cannot handle multi-class problems. We present a novel single-class algorithm called Class Conditional Nearest Neighbor Distribution (CCNND), which mitigates the effects of class imbalance through local geometric structure in the data. Our algorithm can be applied seamlessly to problems with any level of imbalance or number of classes, and new examples are simply added to the training set. We show that it performs as well as or better than top sampling and cost-weighting methods on four imbalanced datasets from the UCI Machine Learning Repository, and then apply it to streaming data from the oil and gas industry alongside a modified nearest neighbor algorithm. Our algorithms competitive performance relative to the state-of-the-art, coupled with its extremely simple implementation and automatic adjustment for minority classes, demonstrates that it is worth further study.

Non-dyadic Haar wavelets for streaming and sensor data

In streaming and sensor data applications, the problems of synopsis construction and outlier detection are important. Due to their low complexity, desirable properties and relative ease of understanding, wavelet based techniques are often used for both synopsis construction and anomaly detection. In streaming data literature, Mallats algorithm [1] is often used to achieve a Haar wavelet decomposition in O(n) time. However, there is one limitation to this popular technique, in that it leads to a dyadic decomposition of data. We demonstrate that the property of non-dyadicity is of considerable use in synopsis construction and anomaly detection. In this regard we present several application results, a synopsis data structure for streaming data that is an order of magnitude superior to the popular Haar based wavelet technique, a method for finding anomalies for sensor data over non-dyadic hierarchies, etc. In our work, we enable non-dyadicity by proposing a Mallat like construction for a wavelet system that admits non-dyadic basis. Our algorithm builds a non-dyadic hierarchical structure, and is more efficient than the state of the art construction. We prove the correctness of our construction by showing that our basis functions demonstrates the properties of a wavelet system.

A Comparison of Statistical Machine Learning Methods in Heartbeat Detection and Classification

In health care, patients with heart problems require quick responsiveness in a clinical setting or in the operating theatre. Towards that end, automated classification of heartbeats is vital as some heartbeat irregularities are time consuming to detect. Therefore, analysis of electro-cardiogram (ECG) signals is an active area of research. The methods proposed in the literature depend on the structure of a heartbeat cycle. In this paper, we use interval and amplitude based features together with a few samples from the ECG signal as a feature vector. We studied a variety of classification algorithms focused especially on a type of arrhythmia known as the ventricular ectopic fibrillation (VEB). We compare the performance of the classifiers against algorithms proposed in the literature and make recommendations regarding features, sampling rate, and choice of the classifier to apply in a real-time clinical setting. The extensive study is based on the MIT-BIH arrhythmia database. Our main contribution is the evaluation of existing classifiers over a range sampling rates, recommendation of a detection methodology to employ in a practical setting, and extend the notion of a mixture of experts to a larger class of algorithms.

Improving customer experience via text mining

Improving customer experience on company web sites is an important aspect of maintaining a competitive edge in the technology industry. To better understand customer behavior, e-commerce sites provide online surveys for individual web site visitors to record their feedback with site performance. This paper describes some areas where text mining appears to have useful applications. For comments from web site visitors, we implemented automated analysis to discover emerging problems on the web site using clustering methods and furthermore devised procedures to assign comments to pre-defined categories using statistical classification. Statistical clustering was based on a Gaussian mixture model and hierarchical clustering to uncover new issues related to customer care-abouts. Statistical classification of comments was studied extensively by applying a variety of popular algorithms. We benchmarked their performance and make some recommendations based on our evaluations.

Markov Random Fields in Pattern Recognition for Semiconductor Manufacturing

Under the most general conditions of an anisotropic Markov random field, we model the two-dimensional spatial distribution of microchips on a silicon wafer. The proposed model improves on its predecessors as it stipulates the spatial correlation of different strengths in all eight directions. Its canonical parameters represent the intensity of failures, main effects, and interactions of neighboring chips. Explicit forms of conditional distributions are derived, and maximum pseudo-likelihood estimates of canonical parameters are obtained. This numerical characteristic summarizes general patterns of clusters of failing chips on a wafer, capturing their size, shape, direction, density, and thickness. It is used to classify incoming wafers to known root-cause categories by matching them to the closest pattern.

Pattern Recognition in Large-Scale Data Sets: Application in Integrated Circuit Manufacturing

It is important in semiconductor manufacturing to identify probable root causes, given a signature. The signature is a vector of electrical test parameters measured on a wafer. Linear discriminant analysis and artificial neural networks are used to classify a signature of test electrical measurements of a failed chip to one of several pre-determined root cause categories. An optimal decision rule that assigns a new incoming signature of a chip to a particular root cause category is employed such that the probability of misclassification is minimized. The problem of classifying patterns with missing data, outliers, collinearity, and non-normality are also addressed. The selected similarity metric in linear discriminant analysis, and the network topology, used in neural networks, result in a small number of misclassifications. An alternative classification scheme is based on the locations of failed chips on a wafer and their spatial dependence. In this case, we model the joint distribution of chips by a Markov random field, estimate its canonical parameters and use them as inputs for the artificial neural network that also classifies the patterns by matching them to the probable root causes.

Modified embedding for multi-regime detection in nonstationary streaming data

Many practical data streams are typically composed of several states known as regimes. In this paper, we invoke phase space reconstruction methods from non-linear time series and dynamical systems for regime detection. But the data collected from sensors is normally noisy, does not have constant amplitude and is sometimes plagued by shifts in the mean. All these aspects make modeling even more difficult. We propose a representation of the time series in the phase space with a modified embedding, which is invariant to translation and scale. The features we use for regime detection are based on comparing trajectory segments in the modified embedding space with cross-correntropy, which is a generalized correlation function. We apply our algorithm to non-linear oscillations, and compare its performance with the standard time delay embedding.

On estimating the mean In a bivariate normal distribution With equal or unequal variances

This paper considers the problem of estimating the mean μx of one of the components of the bivariate normal distribution with equal variances or unequal variances. When the mean of the other component μy is equal to μx it is advantageous to pool the two sample means as an estimator of μx. When the experimenter is uncertain whether μx = μy a preliminary test of significance is used at level α to test μx = μy. Three estimators of μx are considered, (i) preliminary test estimator (PTE), (ii) weighting function estimator (WFE), and (iii) adaptive preliminary test estimator (APTE). The WFE is defined as the linear combination of the two sample means with the weight obtained by minimizing the mean square error. The APTE is a PTE with the weight adopted from WFE. The biases, mean square errors, and relative efficiencies of all the three estimators are studied.

Signature analysis based IC diagnostics-a statistician's perspective

This paper deals with the basic concepts of signature analysis, and will attempt to demonstrate how its implementation would enable efficient utilization of failure analysis engineering resources to analyze field failures and avoid repetitive analyses. This would accomplish the dual objective of improved customer satisfaction and reduced cycle time. Signature analysis methodology can be used in Failure Analysis, Design, Product, and Customer Quality and Reliability Engineering group applications. Starting with definitions, purpose, and various possible scenarios, a formal mathematical framework is developed for computing sample sizes and establishing confidence levels when the failures occur at random or occur in clusters.