41 results about "Regularized least squares" patented technology

The invention discloses an automatic target recognition method of Laplace regularization least square synthetic aperture radar, which relates to radar technology field and aims at improving the discrimination of SAR target image by adopting the method and get a better robustness to directions. The implementation steps in the invention are: firstly to make feature extraction for all the samples in MSTAR database by adopting KPCA, take all the data in training sets as samples with signs, and use the data in training sets as samples without signs to set up a weighted undigraph Gequals to (V, E), see the data point as the peak V of G, then define the similarity of the paired data points as the side of the graph, calculate the Laplacian of the graph and take as a regular term and added to the regularization least square and call as Laplace regularization least square to calculate the relative optimization problem. Then, classify the sample without signs by classifying function gain from exercises. The method can solve the identification problem based on the two-dimension SAR image.

Techniques are disclosed that implement algorithms for rapidly finding the leave-one-out (LOO) error for regularized least squares (RLS) problems over a large number of values of the regularization parameter λ. Algorithms implementing the techniques use approximately the same time and space as training a single regularized least squares classifier / regression algorithm. The techniques include a classification / regression process suitable for moderate sized datasets, based on an eigendecomposition of the unregularized kernel matrix. This process is applied to a number of benchmark datasets, to show empirically that accurate classification / regression can be performed using a Gaussian kernel with surprisingly large values of the bandwidth parameter σ. It is further demonstrated how to exploit this large σ regime to obtain a linear-time algorithm, suitable for large datasets, that computes LOO values and sweeps over λ.

A face recognition method is provided to use sparse representation and regularized least squares-based classification on a computing device. The method includes obtaining an image to be recognized as a test sample y and a set of training images of certain subjects as training sample matrix T, obtaining a sparse representation of the test sample and the training samples including an initial estimation of a sparse vector a, and constructing a new face dictionary comprising training samples with non-zero corresponding coefficients in the sparse vector a for the initial estimation. The method also includes obtaining new coefficients by solving a regularized least squares problem based on the constructed new face dictionary, and determining a face identity of the test sample based on minimum class residual calculated by using the new coefficients.

A simple yet powerful Bayesian model of linear regression is disclosed for methods and systems of machine learning. Unlike previous treatments that have either considered finding hyperparameters through maximum likelihood or have used a simple prior that makes the computation tractable but can lead to overfitting in high dimensions, the disclosed methods use a combination of linear algebra and numerical integration to work a full posterior over hyperparameters in a model with a prior that naturally avoids overfitting. The resulting algorithm is efficient enough to be practically useful. The approach can be viewed as a fully Bayesian version of the discriminative regularized least squares algorithm.

The invention discloses a power equipment fault monitoring method based on a mutual reconstruction single-class auto-encoder.The method comprises the following steps: preprocessing acquired magnetic field information of power equipment in normal operation to obtain a training data sample set, training K mutual reconstruction single-class random auto-encoders WSI-GAE by taking the training data sample set as input to obtain a final encoding result, performing single classification model training by using regularized least square single classification loss, obtaining a fitting error of each trained data sample, and selecting a threshold from fitting error sequences of the data samples arranged from large to small, for newly collected magnetic field information data of the power equipment, the obtained fitting error is compared with a threshold value, and when the fitting error is larger than the threshold value, it can be judged that the power equipment has abnormal conditions such as faults. According to the invention, a single-class classifier technology is used to realize anomaly detection, and the method is more suitable for the target of the invention. The fault abnormity monitoring accuracy of the power equipment is improved.

Techniques are disclosed that implement algorithms for rapidly finding the leave-one-out (LOO) error for regularized least squares (RLS) problems over a large number of values of the regularization parameter λ. Algorithms implementing the techniques use approximately the same time and space as training a single regularized least squares classifier / regression algorithm. The techniques include a classification / regression process suitable for moderate sized datasets, based on an eigendecomposition of the unregularized kernel matrix. This process is applied to a number of benchmark datasets, to show empirically that accurate classification / regression can be performed using a Gaussian kernel with surprisingly large values of the bandwidth parameter σ. It is further demonstrated how to exploit this large σ regime to obtain a linear-time algorithm, suitable for large datasets, that computes LOO values and sweeps over λ.

A method and system use capacity-approaching rateless code to communicate multimedia data even with very short codewords, such as 64 bits or less, via erasure and noise channels. The method provides a way to design the edge degree distribution of rateless codes for any arbitrary channel. Based on an equivalent metric of decoding behavior in any channels, the degree distribution of a rateless code such as Luby-transform codes and raptor codes is optimized based on in-process status of decoding. A regularized least-squares optimization is used to avoid erroneous decoding. Multiple feedbacks can further improve the performance.

The invention discloses a method and device for predicting a drug-target interaction relationship. The method comprises the following steps: introducing drug-target interaction and target-disease interaction to construct a three-layer heterogeneous network of drug-target-disease; constructing a drug similarity matrix and a target similarity matrix based on the three-layer heterogeneous network, wherein the target similarity matrix comprises an inter-target Gaussian kernel similarity matrix and a target-disease Gaussian kernel similarity matrix; calculating a Kronecker product of the drug similarity matrix and the target similarity matrix, and obtaining a prediction result through a regularized least square method; and verifying the prediction result. Compared with a traditional predictionmethod, the method has the advantages that a more complete network structure model is adopted, a more complex similarity matrix space is established, brand-new drug-target interaction is predicted from more perspectives, ultra-scale matrix operation is avoided in the calculation process, and compared with a common FLapRLS method and a common RLS_Kron method, the method provided by the invention has better prediction performance.

Sound discrimination techniques are disclosed that can be employed, for example, in the task of discriminating speech and non-speech in a noisy environment and other noise classification applications. In one particular embodiment, a classifier system is provided that includes a linear Regularized Least Squares classifier used directly on a high-dimensional cortical representation of the target sound. The regularization constant lambda (λ) can be selected automatically, yielding a parameter-free learning system. In addition, the high-dimensional hyperplane can be viewed directly in the cortical space, leading to greater interpretability of the classifier results.

The invention discloses a time-frequency domain modal parameter identification method based on nonlinear frequency modulation component decomposition, comprising the following steps: 1. Obtaining the dynamic response signal of the structure to be identified and setting the sampling time and frequency; 2. Through Fourier The leaf series establishes the redundant Fourier model of the response signal, instantaneous frequency and instantaneous amplitude; 3. Extract the instantaneous frequency information of the response signal through the generalized parameterized time-frequency transformation; 4. According to the obtained instantaneous frequency information, through regularization The instantaneous amplitude information of the response signal is extracted by the least squares method; 5. According to the slope information of the instantaneous frequency and the logarithm of the instantaneous amplitude, the modal parameter identification of the structure is realized by using the linear least squares fitting algorithm; 6. The identification results are analyzed Error Analysis. The invention directly uses the vibration response of the structure for signal analysis and mode parameter identification, is simple and convenient to use, can effectively improve the mode parameter identification accuracy of dense mode structures, and has strong applicability and anti-interference ability.

The invention discloses an image compression method based on active learning and semi-supervised learning, which selects color information of some representative pixel points in an image to learn a model which can predict colors of the remaining pixel points in the image on the basis of the active learning. The image compression method mainly includes the following steps that all pixel points of the image are provided, the image is segmented into a plurality of rectangular areas simply, one pixel point is respectively and randomly selected from each rectangular area, the pixel points are learned actively so as to select the most representative pixel points, grey level values and color values of the most representative pixel points are recorded, grey level values of the remaining pixel points of the image are only recorded, and a compression process is completed. During decompression, Laplace regularized least squares (LapRLS) serving as a semi-supervised learning algorithm is used for predicting and restoring colors of all pixel points.

The present invention provides a cloud detection method and system based on threshold and regularized least squares, including: using the threshold method to perform initial detection on remote sensing images, and obtaining initial cloud data after detection; Classify the above initial cloud data, and obtain the classified cloud data. The cloud detection method and system based on threshold and regularized least squares provided by the present invention first uses the threshold method to perform initial detection on remote sensing images and obtains the initial cloud data after detection, and then uses the regularized least squares method to detect the initial cloud data. Classify the data and obtain the classified cloud data, which greatly improves the accuracy of cloud detection, and, after the initial detection, reduces the number of classes classified using regularized least squares, thereby reducing the computational complexity and improving Computational efficiency of cloud detection.