Showing papers by "Kapil Ahuja published in 2018"

Density-Wise Two Stage Mammogram Classification using Texture Exploiting Descriptors

Abstract: Breast cancer is becoming pervasive with each passing day. Hence, its early detection is a big step in saving the life of any patient. Mammography is a common tool in breast cancer diagnosis. The most important step here is classification of mammogram patches as normal–abnormal and benign–malignant. Texture of a breast in a mammogram patch plays a significant role in these classifications. We propose a variation of Histogram of Gradients (HOG) and Gabor filter combination called Histogram of Oriented Texture (HOT) that exploits this fact. We also revisit the Pass Band - Discrete Cosine Transform (PB-DCT) descriptor that captures texture information well. All features of a mammogram patch may not be useful. Hence, we apply a feature selection technique called Discrimination Potentiality (DP). Our resulting descriptors, DP-HOT and DP-PB-DCT, are compared with the standard descriptors. Density of a mammogram patch is important for classification, and has not been studied exhaustively. The Image Retrieval in Medical Application (IRMA) database from RWTH Aachen, Germany is a standard database that provides mammogram patches, and most researchers have tested their frameworks only on a subset of patches from this database. We apply our two new descriptors on all images of the IRMA database for density wise classification, and compare with the standard descriptors. We achieve higher accuracy than all of the existing standard descriptors (more than 92%).

Localized Multiple Kernel Learning for Anomaly Detection: One-class Classification

Abstract: Multi-kernel learning has been well explored in the recent past and has exhibited promising outcomes for multi-class classification and regression tasks. In this paper, we present a multiple kernel learning approach for the One-class Classification (OCC) task and employ it for anomaly detection. Recently, the basic multi-kernel approach has been proposed to solve the OCC problem, which is simply a convex combination of different kernels with equal weights. This paper proposes a Localized Multiple Kernel learning approach for Anomaly Detection (LMKAD) using OCC, where the weight for each kernel is assigned locally. Proposed LMKAD approach adapts the weight for each kernel using a gating function. The parameters of the gating function and one-class classifier are optimized simultaneously through a two-step optimization process. We present the empirical results of the performance of LMKAD on 25 benchmark datasets from various disciplines. This performance is evaluated against existing Multi Kernel Anomaly Detection (MKAD) algorithm, and four other existing kernel-based one-class classifiers to showcase the credibility of our approach. Our algorithm achieves significantly better Gmean scores while using a lesser number of support vectors compared to MKAD. Friedman test is also performed to verify the statistical significance of the results claimed in this paper.

Scaled and Projected Spectral Clustering with Vector Quantization for Handling Big Data

Abstract: In this modern era, the advent of web technologies and social networking websites is generating a significant amount of data every day. In this scenario, where the data size is now reaching zetta bytes (i.e., 1021), its analysis is very important.Since spectral-based clustering algorithms provide more accurate results than traditional clustering algorithms, we focus on these algorithms. In our work, we propose a modified version of spectral clustering, which we call Projected Spectral Clustering (PSC). As the complexity of the PSC algorithm is Opn3q, where n is the size of the data, we use two variants of vector quantization sampling namely k-Means (KM) and Bisecting k-Means (BKM). To make our algorithm scalable for handling Big Data, we implement it on Apache Spark using two approaches for computing the Gaussian Kernel matrix, which is the most important step here (i.e. Map Reduce and Map Only). We call this algorithm Scalable PSC (SPSC).We measure the accuracy of SPSC using three evaluation criteria tested on a variety of different datasets. Our new algorithm gives good clustering accuracies. Further, we perform another set of experiments on a different number of cores to demonstrate runtime/ scalability efficiency of our algorithm. Finally, we prove this scalability by doing a complexity analysis.

Externalities in Endogenous Sharing Economy Networks.

Abstract: This paper investigates the impact of link formation between a pair of agents on the resource availability of other agents (that is, externalities) in a social cloud network, a special case of endogenous sharing economy networks. Specifically, we study how the closeness between agents and the network size affect externalities. We conjecture, and experimentally support, that for an agent to experience positive externalities, an increase in its closeness is necessary. The condition is not sufficient though. We, then, show that for populated ring networks, one or more agents experience positive externalities due to an increase in the closeness of agents. Further, the initial distance between agents forming a link has a direct bearing on the number of beneficiaries, and the number of beneficiaries is always less than that of non-beneficiaries.

Parallel FPGA Router using Sub-Gradient method and Steiner tree.

Abstract: In the FPGA (Field Programmable Gate Arrays) design flow, one of the most time-consuming step is the routing of nets. Therefore, there is a need to accelerate it. In a recent paper by Hoo et. al., the authors have developed a Linear Programming based framework that parallelizes this routing process to achieve significant speedups (the algorithm is termed as ParaLaR). However, this approach has certain weaknesses. Namely, the constraints violation by the solution and a local minima that could be improved. We address these two issues here. In our paper, we use this framework and solve it using the Primal-Dual sub-gradient method that better exploits the problem properties. We also propose a better way to update the size of the step taken by this iterative algorithm. We perform experiments on a set of standard benchmarks, where we show that our algorithm outperforms the standard existing algorithms (VPR and ParaLaR). We achieve up to 22% improvement in the constraints violation and the standard metric of the minimum channel width when compared with ParaLaR (which is same as in VPR). We achieve about 20% savings in another standard metric of the total wire length (when compared with VPR), which is the same as for ParaLaR. Hence, our algorithm achieves minimum value for all the three parameters. Also, the critical path delay for our algorithm is almost same as compared to VPR and ParaLaR. We achieve relative speedups of 3 times when we run a parallel version of our algorithm using 4 threads.

Stability Analysis of Inexact Solves in Moment Matching based Model Reduction.

Abstract: Recently a new algorithm for model reduction of second order linear dynamical systems with proportional damping, the Adaptive Iterative Rational Global Arnoldi (AIRGA) algorithm (Bonin et. al., 2016), has been proposed. The main computational cost of the AIRGA algorithm is solving a sequence of linear systems. Usually, direct methods (e.g., LU) are used for solving these systems. As model sizes grow, direct methods become prohibitively expensive. Iterative methods (e.g., Krylov) scale well with size, and hence, are a good choice with an appropriate preconditioner. Preconditioned iterative methods introduce errors in linear solves because they are not exact. They solve linear systems up to a certain tolerance. We prove that, under mild conditions, the AIRGA algorithm is backward stable with respect to the errors introduced by these inexact linear solves. Our first assumption is use of a Ritz-Galerkin based solver that satisfies few extra orthogonality conditions. Since Conjugate Gradient (CG) is the most popular method based upon the Ritz-Galerkin theory, we use it. We show how to modify CG to achieve these extra orthogonalities. Modifying CG with the suggested changes is non-trivial. Hence, we demonstrate that using Recycling CG (RCG) helps us achieve these orthogonalities with no code changes. The extra cost of orthogonalizations is often offset by savings because of recycling. Our second and third assumptions involve existence, invertibility and boundedness of two matrices, which are easy to satisfy. While satisfying the backward stability assumptions, by numerical experiments we show that as we iteratively solve the linear systems arising in the AIRGA algorithm more accurately, we obtain a more accurate reduced system. We also show that recycling Krylov subspaces helps satisfy the backward stability assumptions (extra-orthogonalities) at almost no extra cost.

Externalities in Socially-Based Resource Sharing Network.

Abstract: This paper investigates the impact of link formation between a pair of agents on resource availability of other agents in a social cloud network, which is a special case of socially-based resource sharing systems. Specifically, we study the correlation between externalities, network size, and network density. We first conjecture and experimentally support that if an agent experiences positive externalities, then its closeness (harmonic centrality measure) should increase. Next, we show the following for ring networks: in less populated networks no agent experiences positive externalities; in more populated networks a set of agents experience positive externalities, and larger the distance between agents forming a link, more the number of beneficiaries; and the number of beneficiaries is always less than the number of non-beneficiaries. Finally, we show that network density is inversely proportional to positive externalities, and further, it plays a crucial role in determining the kind of externalities.

Vector Quantized Spectral Clustering applied to Soybean Whole Genome Sequences.

Abstract: We develop a Vector Quantized Spectral Clustering (VQSC) algorithm that is a combination of Spectral Clustering (SC) and Vector Quantization (VQ) sampling for grouping Soybean genomes. The inspiration here is to use SC for its accuracy and VQ to make the algorithm computationally cheap (the complexity of SC is cubic in-terms of the input size). Although the combination of SC and VQ is not new, the novelty of our work is in developing the crucial similarity matrix in SC as well as use of k-medoids in VQ, both adapted for the Soybean genome data. We compare our approach with commonly used techniques like UPGMA (Un-weighted Pair Graph Method with Arithmetic Mean) and NJ (Neighbour Joining). Experimental results show that our approach outperforms both these techniques significantly in terms of cluster quality (up to 25% better cluster quality) and time complexity (order of magnitude faster).

Preconditioned Linear Solves for Parametric Model Order Reduction.

Abstract: The main computational cost of algorithms for computing reduced-order models of parametric dynamical systems is in solving sequences of very large and sparse linear systems We focus on efficiently solving these linear systems, arising while reducing second-order linear dynamical systems, by iterative methods with appropriate preconditioners We propose that the choice of underlying iterative solver is problem dependent We propose the use of block variant of the underlying iterative method because often all right-hand-side are available together Since, Sparse Approximate Inverse (SPAI) preconditioner is a general preconditioner that can be naturally parallelized, we propose its use Our most novel contribution is a technique to cheaply update the SPAI preconditioner, while solving the parametrically changing linear systems We support our proposed theory by numerical experiments where we first show benefit of 80% in time by using a block iterative method, and a benefit of 70% in time by using SPAI updates

Data backup network formation with heterogeneous agents: (Extended abstract)

Abstract: Social storage systems [1], [2], [3] are becoming increasingly popular compared to the existing data backup systems like local, centralized and P2P systems. An endogenously built symmetric social storage model and its aspects like the utility of each agent, bilateral stability, contentment, and efficiency have been extensively discussed in [4]. We include heterogeneity in this model by using the concept of Social Range Matrix from [5]. Now, each agent is concerned about its perceived utility, which is a linear combination of its utility as well as others utilities (depending upon whether the pair are friends, enemies or do not care about each other). We derive conditions when two agents may want to add or delete a link, and provide an algorithm that checks if a bilaterally stable network is possible or not. Finally, we take some special Social Range Matrices and prove that under certain conditions on network parameters, a bilaterally stable network is unique.