Sas Institute Inc.

Arin Chaudhuri

Deovrat Vijay Kakde

Sergiy Peredriy

Hansi Jiang

Wei Xiao

Jorge Manuel Gomes Da Silva

Yuwei Liao

Seung Hyun Kong

Anya Mary McGuirk

Saba Emrani

Haoyu Wang

Carol Wagih Sadek

Maria Jahja

Vincent L Deters

Kevin L Scott

Byron Davis Biggs

Allen Joseph Langlois

Kai Shen

Yu Liang

Zekun Xu

A heart-rate detection system can receive heartbeat data generated by a wearable heart-rate sensor worn by a wearer. The system can then execute a noise-reduction process for reducing noise in the heartbeat data. The noise-reduction process can involve applying a lowpass filter to the heartbeat data, generating wavelet coefficients by applying a wavelet transform to the filtered heartbeat data, and generating a reduced set of wavelet coefficients by thresholding the wavelet coefficients. An inverse wavelet signal can then be generated by applying an inverse wavelet transform to the reduced set of wavelet coefficients. R-peaks can be identified by performing peak detection on the instantaneous amplitudes of the data points in the inverse wavelet signal. A heart rate curve can then be generated based on the R-peaks and modified by applying a Hampel filter. Heartbeat data can then be generated based on the modified heart rate curve for output.

Robust heart-rate detection techniques for wearable heart-rate sensors

Anomalies in a target object can be detected and diagnosed using improved Mahalanobis-Taguchi system (MTS) techniques. For example, an anomaly detection and diagnosis (ADD) system can receive a set of measurements associated with attributes of a target object. A Mahalanobis distance (MD) can be determined using a generalized inverse matrix. An abnormal condition can be detected when the MD is greater than a predetermined threshold value. The ADD system can determine an importance score for each measurement of a corresponding attribute. The attribute whose measurement has the highest importance score can be determined to be responsible for the abnormal condition.

Anomaly detection and diagnostics based on multivariate analysis

A singular value decomposition (SVD) is computed of a first matrix to define a left matrix, a diagonal matrix, and a right matrix. The left matrix, the diagonal matrix, and the right matrix are updated using an arrowhead matrix structure defined from the diagonal matrix and by adding a next observation vector to a last row of the first matrix. The updated left matrix, the updated diagonal matrix, and the updated right matrix are updated using a diagonal-plus-rank-one (DPR1) matrix structure defined from the updated diagonal matrix and by removing an observation vector from a first row of the first matrix. Eigenpairs of the DPR1 matrix are computed based on whether a value computed from the updated left matrix is positive or negative. The left matrix updated in (C), the diagonal matrix updated in (C), and the right matrix updated in (C) are output.

Incremental singular value decomposition in support of machine learning

A computer transforms high-dimensional data into low-dimensional data. (A) A distance matrix is computed from observation vectors. (B) A kernel matrix is computed from the distance matrix using a bandwidth value. (C) The kernel matrix is decomposed using an eigen decomposition to define eigenvalues. (D) A predefined number of largest eigenvalues are selected from the eigenvalues. (E) The selected largest eigenvalues are summed. (F) A next bandwidth value is computed based on the summed eigenvalues. (A) through (F) are repeated with the next bandwidth value until a stop criterion is satisfied. Each observation vector of the observation vectors is transformed into a second space using a kernel principal component analysis with the next bandwidth value and the kernel matrix. The second space has a dimension defined by the predefined number of first eigenvalues. Each transformed observation vector is output.

High dimensional to low dimensional data transformation and visualization system

A computing system determines if an event has occurred. A first window is defined that includes a subset of a plurality of observation vectors modeled as an output of an autoregressive causal system. A magnitude adjustment vector is computed from a mean computed for a matrix of magnitude values that includes a column for each window of a plurality of windows. The first window is stored in a next column of the matrix of magnitude values. Each cell of the matrix of magnitude values includes an estimated power spectrum value for a respective window and a respective frequency. A second matrix of magnitude values is updated using the magnitude adjustment vector. Each cell of the second matrix of magnitude values includes an adjusted power spectrum value for the respective window and the respective frequency. A peak is detected from the next column of the second matrix of magnitude values.

Data monitoring system

A computer transforms high-dimensional data into low-dimensional data. A distance is computed between a selected observation vector and each observation vector of a plurality of observation vectors, a nearest neighbors are selected using the computed distances, and a first sigmoid function is applied to compute a distance similarity value between the selected observation vector and each of the selected nearest neighbors where each of the computed distance similarity values is added to a first matrix. The process is repeated with each observation vector of the plurality of observation vectors as the selected observation vector. An optimization method is executed with an initial matrix, the first matrix, and a gradient of a second sigmoid function that computes a second distance similarity value between the selected observation vector and each of the nearest neighbors to transform each observation vector of the plurality of observation vectors into the low-dimensional space.

A computing device determines hyperparameter values for outlier detection. An LOF score is computed for observation vectors using a neighborhood size value. Outlier observation vectors are selected from the observation vectors. Outlier mean and outlier variance values are computed of the LOF scores of the outlier observation vectors. Inlier observation vectors are selected from the observation vectors that have highest computed LOF scores of the observation vectors that are not included in the outlier observation vectors. Inlier mean and inlier variance values are computed of the LOF scores of the inlier observation vectors. A difference value is computed using the outlier mean and variance values and the inlier mean and variance values. The process is repeated with each neighborhood size value of a plurality of neighborhood size values. A tuned neighborhood size value is selected as the neighborhood size value associated with an extremum value of the difference value.

Local outlier factor hyperparameter tuning for data outlier detection

A computing device determines a bandwidth parameter value for outlier detection or data classification. A mean pairwise distance value is computed between observation vectors. A tolerance value is computed based on a number of observation vectors. A scaling factor value is computed based on a number of observation vectors and the tolerance value. A Gaussian bandwidth parameter value is computed using the mean pairwise distance value and the scaling factor value. An optimal value of an objective function is computed that includes a Gaussian kernel function that uses the computed Gaussian bandwidth parameter value. The objective function defines a support vector data description model using the observation vectors to define a set of support vectors. The Gaussian bandwidth parameter value and the set of support vectors are output for determining if a new observation vector is an outlier or for classifying the new observation vector.

Bandwidth selection in support vector data description for classification or outlier detection

A computing device updates an estimate of one or more principal components for a next observation vector. An initial observation matrix is defined with first observation vectors. A number of the first observation vectors is a predefined window length. Each observation vector of the first observation vectors includes a plurality of values. A principal components decomposition is computed using the initial observation matrix. The principal components decomposition includes a sparse noise vector s, a first singular value decomposition vector U, and a second singular value decomposition vector v for each observation vector of the first observation vectors. A rank r is determined based on the principal components decomposition. A next principal components decomposition is computed for a next observation vector using the determined rank r. The next principal components decomposition is output for the next observation vector and monitored to determine a status of a physical object.

Monitoring, detection, and surveillance system using principal component analysis with machine and sensor data

A computing device detects an abnormal observation vector using a principal components decomposition. The principal components decomposition includes a sparse noise vector scomputed for the observation vector that includes a plurality of values, wherein each value is associated with a variable to define a plurality of variables. The sparse noise vector shas a dimension equal to m a number of the plurality of variables. A zero counter time series value &#x109;is computed using &#x109;=&#x3a3;s[i]. A probability value for &#x109;is computed using p=&#x3a3;H[i]/&#x3a3;H[i], where H[i] includes a count of a number of times each value of &#x109;occurred for previous observation vectors. The probability value is compared with a predefined abnormal observation probability value. An abnormal observation indicator is set when the probability value indicates the observation vector is abnormal. The observation vector is output when the probability value indicates the observation vector is abnormal.

A computing device determines a kernel parameter value for a support vector data description for outlier identification. A first candidate optimal kernel parameter value is computed by computing a first optimal value of a first objective function that includes a kernel function for each of a plurality of kernel parameter values from a starting kernel parameter value to an ending kernel parameter value using an incremental kernel parameter value. The first objective function is defined for a SVDD model using observation vectors to define support vectors. A number of the observation vectors is a predefined sample size. The predefined sample size is incremented by adding a sample size increment. A next candidate optimal kernel parameter value is computed with an incremented number of vectors until a computed difference value is less than or equal to a predefined convergence value.

Kernel parameter selection in support vector data description for outlier identification

A computing device determines an SVDD to identify an outlier in a dataset. First and second sets of observation vectors of a predefined sample size are randomly selected from a training dataset. First and second optimal values are computed using the first and second observation vectors to define a first set of support vectors and a second set of support vectors. A third optimal value is computed using the first set of support vectors updated to include the second set of support vectors to define a third set of support vectors. Whether or not a stop condition is satisfied is determined by comparing a computed value to a stop criterion. When the stop condition is not satisfied, the first set of support vectors is defined as the third set of support vectors, and operations are repeated until the stop condition is satisfied. The third set of support vectors is output.

Fast training of support vector data description using sampling

A computing device identifies outliers. Support vectors, Lagrange constants, a center threshold value, an upper control limit value, and a lower control limit value are received that define a normal operating condition of a system. The center threshold value, the upper control limit value, and the lower control limit value are computed from the vectors and the Lagrange constants. A first plurality of observation vectors is received for a predefined window length. A window threshold value and a window center vector are computed. A window distance value is computed between the window center vector and the support vectors. Based on comparisons between the computed values and the received values, the first plurality of observation vectors is identified as an outlier relative to the normal operating condition of the system. When the first plurality of observation vectors are identified as the outlier, an alert is output.

Monitoring system based on a support vector data description

A computer-readable medium is configured to determine a support vector data description (SVDD). For each of a plurality of values for a kernel parameter, an optimal value of an objective function defined for an SVDD model using a kernel function, a read plurality of data points, and a respective value for the kernel parameter is computed to define a plurality of sets of support vectors. A plurality of first derivative values are computed for the objective function as a difference between the computed optimal values associated with successive values for the kernel parameter. A plurality of second derivative values are computed for the objective function as a difference between the computed plurality of first derivative values associated with successive values for the kernel parameter. A kernel parameter value is selected where the computed plurality of second derivative values first exceeds zero.

Growing community of inventors

Raleigh, NC, United States of America

Arin Chaudhuri