Hybrid Working Set Algorithm for SVM Learning With a Kernel Coprocessor on FPGA

Support vector machines (SVM) are a popular class of supervised models in machine learning. The associated compute intensive learning algorithm limits their use in real-time applications. This paper presents a fully scalable architecture of a coprocessor, which can compute multiple rows of the kernel matrix in parallel. Further, we propose an extended variant of the popular decomposition technique, sequential minimal optimization, which we call hybrid working set (HWS) algorithm, to effectively utilize the benefits of cached kernel columns and the parallel computational power of the coprocessor. The coprocessor is implemented on Xilinx Virtex 7 field-programmable gate array-based VC707 board and achieves a speedup of upto 25x for kernel computation over single threaded computation on Intel Core i5. An application speedup of upto 15x over software implementation of LIBSVM and speedup of upto 23x over SVMLight is achieved using the HWS algorithm in unison with the coprocessor. The reduction in the number of iterations and sensitivity of the optimization time to variation in cache size using the HWS algorithm are also shown.

Sparse Representation based Anomaly Detection using HOMV in H.264 Compressed Videos

In this paper, we have proposed an anomaly detection algorithm based on Histogram of Oriented Motion Vectors (HOMV) 1] in sparse representation framework. Usual behavior is learned at each location by sparsely representing the HOMVs over learnt normal feature bases obtained using an online dictionary learning algorithm. In the end, anomaly is detected based on the likelihood of the occurrence of sparse coefficients at that location. The proposed approach is found to be robust compared to existing methods as demonstrated in the experiments on UCSD Ped1 and UCSD Ped2 datasets.

Joint Bilateral Filtering based Non-local Means Image Denoising

This paper proposes a denoising algorithm which performs non-local means bilateral filtering. As existing literature suggests, non-local means (NLM) is one of the widely used denoising techniques, but has a critical drawback of smoothing of edges. In order to improve this, we perform fast and efficient NLM using Approximate Nearest Neighbour Fields and improve the edge content in denoising by formulating a joint-bilateral filter. Using the proposed joint bilateral, we are able to denoise smooth regions using the NLM approach and efficient edge reconstruction is obtained from the bilateral filter. Furthermore, to avoid tedious parameter selection, we carry out a noise estimation before performing joint bilateral filtering. The proposed approach is observed to perform well on high noise images.

Entropy Constrained Exemplar-based Image Inpainting

Image inpainting is the process of filling the unwanted region in an image marked by the user. It is used for restoring old paintings and photographs, removal of red eyes from pictures, etc. In this paper, we propose an efficient inpainting algorithm which takes care of false edge propagation. We use the classical exemplar based technique to find out the priority term for each patch. To ensure that the edge content of the nearest neighbor patch found by minimizing L-2 distance between patches, we impose an additional constraint that the entropy of the patches be similar. Entropy of the patch acts as a good measure of edge content. Additionally, we fill the image by considering overlapping patches to ensure smoothness in the output. We use structural similarity index as the measure of similarity between ground truth and inpainted image. The results of the proposed approach on a number of examples on real and synthetic images show the effectiveness of our algorithm in removing objects and thin scratches or text written on image. It is also shown that the proposed approach is robust to the shape of the manually selected target. Our results compare favorably to those obtained by existing techniques

A Design for Performance Improvement of Clock Synchronization in WSNs using a TDMA-based MAC Protocol

Clock synchronization in a wireless sensor network (WSN) is quite essential as it provides a consistent and a coherent time frame for all the nodes across the network. Typically, clock synchronization is achieved by message passing using a contention-based scheme for media access, like carrier sense multiple access (CSMA). The nodes try to synchronize with each other, by sending synchronization request messages. If many nodes try to send messages simultaneously, contention-based schemes cannot efficiently avoid collisions. In such a situation, there are chances of collisions, and hence, message losses, which, in turn, affects the convergence of the synchronization algorithms. However, the number of collisions can be reduced with a frame based approach like time division multiple access (TDMA) for message passing. In this paper, we propose a design to utilize TDMA-based media access and control (MAC) protocol for the performance improvement of clock synchronization protocols. The basic idea is to use TDMA-based transmissions when the degree of synchronization improves among the sensor nodes during the execution of the clock synchronization algorithm. The design significantly reduces the collisions among the synchronization protocol messages. We have simulated the proposed protocol in Castalia network simulator. The simulation results show that the proposed protocol significantly reduces the time required for synchronization and also improves the accuracy of the synchronization algorithm.

Ray tracing based path-length calculations for polarized light tomographic imaging

A ray tracing based path length calculation is investigated for polarized light transport in a pixel space. Tomographic imaging using polarized light transport is promising for applications in optical projection tomography of small animal imaging and turbid media with low scattering. Polarized light transport through a medium can have complex effects due to interactions such as optical rotation of linearly polarized light, birefringence, diattenuation and interior refraction. Here we investigate the effects of refraction of polarized light in a non-scattering medium. This step is used to obtain the initial absorption estimate. This estimate can be used as prior in Monte Carlo (MC) program that simulates the transport of polarized light through a scattering medium to assist in faster convergence of the final estimate. The reflectance for p-polarized (parallel) and s-polarized (perpendicular) are different and hence there is a difference in the intensities that reach the detector end. The algorithm computes the length of the ray in each pixel along the refracted path and this is used to build the weight matrix. This weight matrix with corrected ray path length and the resultant intensity reaching the detector for each ray is used in the algebraic reconstruction (ART) method. The proposed method is tested with numerical phantoms for various noise levels. The refraction errors due to regions of different refractive index are discussed, the difference in intensities with polarization is considered. The improvements in reconstruction using the correction so applied is presented. This is achieved by tracking the path of the ray as well as the intensity of the ray as it traverses through the medium.

DETERMINISTIC ALGORITHMS FOR MATCHING AND PACKING PROBLEMS BASED ON REPRESENTATIVE SETS

In this work, we study the well-known r-DIMENSIONAL k-MATCHING ((r, k)-DM), and r-SET k-PACKING ((r, k)-SP) problems. Given a universe U := U-1 ... U-r and an r-uniform family F subset of U-1 x ... x U-r, the (r, k)-DM problem asks if F admits a collection of k mutually disjoint sets. Given a universe U and an r-uniform family F subset of 2(U), the (r, k)-SP problem asks if F admits a collection of k mutually disjoint sets. We employ techniques based on dynamic programming and representative families. This leads to a deterministic algorithm with running time O(2.851((r-1)k) .vertical bar F vertical bar. n log(2)n . logW) for the weighted version of (r, k)-DM, where W is the maximum weight in the input, and a deterministic algorithm with running time O(2.851((r-0.5501)k).vertical bar F vertical bar.n log(2) n . logW) for the weighted version of (r, k)-SP. Thus, we significantly improve the previous best known deterministic running times for (r, k)-DM and (r, k)-SP and the previous best known running times for their weighted versions. We rely on structural properties of (r, k)-DM and (r, k)-SP to develop algorithms that are faster than those that can be obtained by a standard use of representative sets. Incorporating the principles of iterative expansion, we obtain a better algorithm for (3, k)-DM, running in time O(2.004(3k).vertical bar F vertical bar . n log(2)n). We believe that this algorithm demonstrates an interesting application of representative families in conjunction with more traditional techniques. Furthermore, we present kernels of size O(e(r)r(k-1)(r) logW) for the weighted versions of (r, k)-DM and (r, k)-SP, improving the previous best known kernels of size O(r!r(k-1)(r) logW) for these problems.

Empirical Analysis of Sampling Based Estimators for Evaluating RBMs

The Restricted Boltzmann Machines (RBM) can be used either as classifiers or as generative models. The quality of the generative RBM is measured through the average log-likelihood on test data. Due to the high computational complexity of evaluating the partition function, exact calculation of test log-likelihood is very difficult. In recent years some estimation methods are suggested for approximate computation of test log-likelihood. In this paper we present an empirical comparison of the main estimation methods, namely, the AIS algorithm for estimating the partition function, the CSL method for directly estimating the log-likelihood, and the RAISE algorithm that combines these two ideas.

Spatio-temporal Map Formation Based on a Potential Function

We revisit the problem of temporal self organization using activity diffusion based on the neural gas (NGAS) algorithm. Using a potential function formulation motivated by a spatio-temporal metric, we derive an adaptation rule for dynamic vector quantization of data. Simulations results show that our algorithm learns the input distribution and time correlation much faster compared to the static neural gas method over the same data sequence under similar training conditions.

Comparison of Matching Pursuit Algorithm with Other Signal Processing Techniques for Computation of the Time-Frequency Power Spectrum of Brain Signals

Signals recorded from the brain often show rhythmic patterns at different frequencies, which are tightly coupled to the external stimuli as well as the internal state of the subject. In addition, these signals have very transient structures related to spiking or sudden onset of a stimulus, which have durations not exceeding tens of milliseconds. Further, brain signals are highly nonstationary because both behavioral state and external stimuli can change on a short time scale. It is therefore essential to study brain signals using techniques that can represent both rhythmic and transient components of the signal, something not always possible using standard signal processing techniques such as short time fourier transform, multitaper method, wavelet transform, or Hilbert transform. In this review, we describe a multiscale decomposition technique based on an over-complete dictionary called matching pursuit (MP), and show that it is able to capture both a sharp stimulus-onset transient and a sustained gamma rhythm in local field potential recorded from the primary visual cortex. We compare the performance of MP with other techniques and discuss its advantages and limitations. Data and codes for generating all time-frequency power spectra are provided.

Parameters Inversion of Fluid-Saturated Porous Media Based on Neural Networks

The multi-layers feedforward neural network is used for inversion of material constants of fluid-saturated porous media. The direct analysis of fluid-saturated porous media is carried out with the boundary element method. The dynamic displacement responses obtained from direct analysis for prescribed material parameters constitute the sample sets training neural network. By virtue of the effective L-M training algorithm and the Tikhonov regularization method as well as the GCV method for an appropriate selection of regularization parameter, the inverse mapping from dynamic displacement responses to material constants is performed. Numerical examples demonstrate the validity of the neural network method.

Simulating shock to detonation transition: Algorithm and results

An algorithm based on flux-corrected transport and the Lagrangian finite element method is presented for solving the problem of shock dynamics. It is verified through the model problem of one-dimensional strain elastoplastic shock wave propagation that the algorithm leads to stable, non-oscillatory results. Shock initiation and detonation wave propagation is simulated using the algorithm, and some interesting results are obtained. (C) 1999 Academic Press.

A Bayesian wavelet-based multidimensional deconvolution with sub-band emphasis

This paper proposes a new algorithm for waveletbased multidimensional image deconvolution which employs subband-dependent minimization and the dual-tree complex wavelet transform in an iterative Bayesian framework. In addition, this algorithm employs a new prior instead of the popular ℓ1 norm, and is thus able to embed a learning scheme during the iteration which helps it to achieve better deconvolution results and faster convergence. © 2008 IEEE.

A species conserving genetic algorithm for multimodal function optimization

This paper introduces a new technique called species conservation for evolving parallel subpopulations. The technique is based on the concept of dividing the population into several species according to their similarity. Each of these species is built around a dominating individual called the species seed. Species seeds found in the current generation are saved (conserved) by moving them into the next generation. Our technique has proved to be very effective in finding multiple solutions of multimodal optimization problems. We demonstrate this by applying it to a set of test problems, including some problems known to be deceptive to genetic algorithms.

Simulation-based optimal sensor scheduling with application to observer trajectory planning

The sensor scheduling problem can be formulated as a controlled hidden Markov model and this paper solves the problem when the state, observation and action spaces are continuous. This general case is important as it is the natural framework for many applications. The aim is to minimise the variance of the estimation error of the hidden state w.r.t. the action sequence. We present a novel simulation-based method that uses a stochastic gradient algorithm to find optimal actions. © 2007 Elsevier Ltd. All rights reserved.