Frontiers in Signal Processing

18

Authors: Seiichi Nakamori

This paper proposes the robust recursive least-squares (RLS) Wiener fixed-point smoother and filter in linear discrete-time stochastic systems with parameter uncertainties. The uncertain parameters exist in the observation matrix and the system matrix. The uncertain parameters cause to generate the degraded signal. In this paper, the degraded signal process is fitted to the finite order autoregressive (AR) model. The robust RLS Wiener estimators use the system matrices and the observation matrices for both the signal and the degraded signal, the variance of the state vector for the degraded signal, the cross-variance of the state vector for the signal with the state vector for the degraded signal, and the variance of the white observation noise. Also, this paper proposes the robust recursive fixed-point smoother and filter, by using the covariance information of the state vector for the degraded signal, the cross-covariance information of the state vector for the signal with the state vector for the degraded signal, the observation matrices for both the degraded signal and the signal besides the variance of the white observation noise. In estimating the signal process expressed by the second order AR model, the proposed robust RLS Wiener filter is superior in estimation accuracy to the robust Kalman filter and the RLS Wiener filter.

16

Authors: Zhao Dongbo, Li Hui

Least squares support vector machine (LS-SVM) has a large amount of computation and sparsity. Aiming at this problem, a fast sparse approximation least squares support vector machine (FSALS-SVM) algorithm is proposed. The algorithm uses an iterative algorithm of complexity to ac-celerate the calculation of the inverse of the kernel matrix and the sparse processing of the support vector machine through the pruning algorithm, thus reducing the computational complexity. The classification and recognition experiments of one dimension range profile of one dimensional radar target show that the FSALS-SVM is more sparsely under the same generalization performance.

29

Authors: Yuriy S. Shmaliy, Yrjö Neuvo, Sanowar Khan

Extracting an estimate of a slowly varying signal corrupted by noise is a common task. Examples can be found in industrial, scientific and biomedical instrumentation. Depending on the nature of the application the signal estimate is allowed to be a delayed estimate of the original signal or, in the other extreme, no delay is tolerated. These cases are commonly referred to as filtering, prediction, and smoothing depending on the amount of advance or lag between the input data set and the output data set. In this review paper we provide a comprehensive set of design and analysis tools for designing unbiased FIR filters, predictors, and smoothers for slowly varying signals, i.e. signals that can be modeled by low order polynomials. Explicit expressions of parameters needed in practical implementations are given. Real life examples are provided including cases where the method is extended to signals that are piecewise slowly varying. A critical view on recursive implementations of the algorithms is provided.

5

Authors: Narek Abroyan

During the last several years machine learning started to revolutionize many industrial fields by replacing human intellectual work with recent technologies. Machine learning has started to be used in financial sphere as well for predicting stock prices, detecting fraud actions etc. In this work, we are focusing on financial market risk classification, which is a part of fraud action detection problem. Although artificial intelligence researchers and specialists achieved notable results in visual, voice signal and natural language processing tasks by using new methods and approaches of deep learning, such as convolutional and recurrent neural networks, not many results are in the sphere of elaboration of real-time non-stationary data, such as financial data. Moreover, methods which are used in industry usually are not published. The goal of this work is exploring, experimenting and providing new and more effective methods of classification of financial non-stationary risk data by using neural networks.

14

Authors: Sarkis Agaian, John T. Gill III

The Golden ratio has played an important role in musical composition, architecture, visual art, science, and increasingly in signal processing [1,2,3]. Underlying many of these applications are several extensions of the golden proportions including the Golden p-Section by Stakhov, the generalized Golden section by Bradley, and others [4,5]. In this paper we review and introduce generalizations of the Golden ratio. We show that there exists a fundamental connection between the limit of two consecutive terms of recursive sequences, the generalized (p, q)-Golden ratio and the Golden ratio generated by the characteristic equation. We apply these generalizations to forecasting financial time series to illustrate one of their applications in signal processing.

7

Authors: Nematollah Zarmehi, Mohammad Javad Barikbin

This paper proposes a robust watermarking method for uncompressed video data against H.264/AVC and H.265/HEVC compression standards. We embed the watermark data in the midrange transform coefficients of a block that is less similar to its corresponding block in the previous and next frames. This idea makes the watermark robust against the compression standards that use the inter prediction technique. The last two video compression standards also use inter prediction for motion compensation like previous video compression standards. Therefore, the proposed method is also well suited with these standards. Simulation results show the adequate robustness and transparency of our watermarking scheme.

8

Authors: Mantosh Biswas, Hari Om

This paper proposes a hybrid image denoising technique using neighbouring wavelet coefficients. The NeighShrink method groups the wavelet coefficients in non overlapping blocks and then thresholds empirically them blockwise. This method does not give good quality of image since it removes too many small wavelet coefficients. Our proposed scheme retains the modified coefficients and also gives good performance in terms of peak signal-to-noise ratio.

17

Authors: Xiaolong Cui, Taieb Znati, Rami Melhem

With the concerted efforts from researchers in hardware, software, algorithm, and data management, HPC is moving towards extreme-scale, featuring a computing capability of quintillion (1018) FLOPS. As we approach the new era of computing, however, several daunting scalability challenges remain to be conquered. Delivering extreme-scale performance will require a computing platform that supports billion-way parallelism, necessitating a dramatic increase in the number of computing, storage, and networking components. At such a large scale, failure would become a norm rather than an exception, driving the system to significantly lower efficiency with unprecedented amount of power consumption. To tackle this challenge, we propose an adaptive and power-aware algorithm, referred to as Lazy Shadowing, as an efficient and scalable approach to achieve high-levels of resilience, through forward progress, in extreme-scale, failure-prone computing environments. Lazy Shadowing associates with each process a “shadow” (process) that executes at a reduced rate, and opportunistically rolls forward each shadow to catch up with the its leading process during failure recovery. Compared to existing fault tolerance methods, our approach can achieve 20% energy saving with potential reduction in solution time at scale.