Possible solutions to the problem of combining classifiers can be divided into three categories according levels information available from various classifiers. Four approaches based on different methodologies are proposed for solving this problem. One is suitable individual such as Bayesian, k-nearest-neighbor, and distance The other could used any kind On applying these methods combine several recognizing totally unconstrained handwritten numerals, experimental results show that...

It is shown that frequency sensitive competitive learning (FSCL), one version of the recently improved (CL) algorithms, significantly deteriorates in performance when number units inappropriately selected. An algorithm called rival penalized (RPCL) proposed. In this algorithm, not only winner unit modified to adapt input for each input, but its (the 2nd winner) delearned by a smaller rate. RPCL can be regarded as an unsupervised extension Kohonen's supervised LVQ2. has ability automatically...

The effectiveness of the treatment breast cancer depends on its timely detection. An early step in diagnosis is cytological examination material obtained directly from tumor. This work reports advances computer-aided based analysis images fine needle biopsies to characterize these as either benign or malignant. Instead relying accurate segmentation cell nuclei, nuclei are estimated by circles using circular Hough transform. resulting then filtered keep only high-quality estimations for...

Principal curves have been defined as "self-consistent" smooth which pass through the "middle" of a d-dimensional probability distribution or data cloud. They give summary and also serve an efficient feature extraction tool. We take new approach by defining principal continuous given length minimize expected squared distance between curve points space randomly chosen according to distribution. The definition makes it possible theoretically analyze learning from training leads practical...

Two results are presented concerning the consistency of $k$-nearest neighbor regression estimate. We show that all modes convergence in $L_1$ (in probability, almost sure, complete) equivalent if variable is bounded. Under additional conditional $k/\log n \rightarrow \infty$ we also obtain strong universal

Training a support vector machine on data set of huge size with thousands classes is challenging problem. This paper proposes an efficient algorithm to solve this The key idea introduce parallel optimization step quickly remove most the nonsupport vectors, where block diagonal matrices are used approximate original kernel matrix so that problem can be split into hundreds subproblems which solved more efficiently. In addition, some effective strategies such as caching and computation...

Proposes an algorithm to find piecewise linear skeletons of handwritten characters by using principal curves. The development the method was inspired apparent similarity between definition curves (smooth which pass through "middle" a cloud points) and medial axes that run equidistantly from contours character image). central fitting-and-smoothing step is extension polygonal line algorithm, approximates data sets extended graphs complemented with two steps specific task skeletonization:...

An estimate $\sum^n_{i=1} Y_iK((x - X_i)/h)/\sum^n_{j=1} K((x X_j)/h)$, calculated from a sequence $(X_1, Y_1), \cdots, (X_n, Y_n)$ of independent pairs random variables distributed as pair $(X, Y)$, converges to the regression $E\{Y\mid X = x\}$ $n$ tends infinity in probability for almost all $(\mu) x \in R^d$, provided that $E|Y| < \infty, h \rightarrow 0$ and $nh^d \infty$ $n \infty$. The result is true distributions $\mu$ $X$. If, moreover, $|Y| \leq \gamma $nh^d/\log n \infty$,...

Pricing of American options in discrete time is considered, where the option allowed to be based on several underlyings. It assumed that price processes underlyings are given Markov processes. We use Monte Carlo approach generate artificial sample paths these processes, and then we least squares neural networks regression estimates estimate from this data so-called continuation values, which defined as mean values for at t subject constraint not exercised t. Results concerning consistency...

In this paper, we introduce the so-called hierarchical interaction models, where assume that computation of value a function m : ℝ <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">d</sup> → is done in several layers, each layer at most d* inputs computed by previous evaluated. We investigate two different regression estimates based on polynomial splines and neural networks, show if satisfies model all occurring functions are smooth, rate...

The denoising of a natural image corrupted by Gaussian noise is classical problem in signal or processing. Donoho and his coworkers at Stanford pioneered wavelet scheme thresholding the coefficients arising from standard discrete transform. This work has been widely used science engineering applications. However, this tends to kill too many that might contain useful information. In paper, we propose one incorporating neighbouring coefficients, namely NeighShrink. approach valid because large...

This correspondence presents a segmentation and fitting method using new robust estimation technique. We present with high breakdown point which can tolerate more than 80% of outliers. The randomly samples appropriate range image points in the current processing region solves equations determined by these for parameters selected primitive type. From K samples, we choose one set sample that determines best-fit equation largest homogeneous surface patch region. choice is made measuring...

Classification of Breast Cancer Malignancy Using Cytological Images Fine Needle Aspiration Biopsies According to the World Health Organization (WHO), breast cancer (BC) is one most deadly cancers diagnosed among middle-aged women. Precise diagnosis and prognosis are crucial reduce high death rate. In this paper we present a framework for automatic malignancy grading fine needle aspiration biopsy tissue. The grade important factors taken into consideration during prediction behavior after...

Deep neural networks (DNNs) achieve impressive results for complicated tasks like object detection on images and speech recognition. Motivated by this practical success, there is now a strong interest in showing good theoretical properties of DNNs. To describe which DNNs perform well when they fail, it key challenge to understand their performance. The aim paper contribute the current statistical theory We apply high dimensional data we show that least squares regression estimates using are...

We apply the method of complexity regularization to derive estimation bounds for nonlinear function using a single hidden layer radial basis network. Our approach differs from previous neural-network learning schemes in that we operate with random covering numbers and l/sub 1/ metric entropy, making it possible consider much broader families activation functions, namely functions bounded variation. Some constraints previously imposed on network parameters are also eliminated this way. The is...