UMD Theses and Dissertations

Permanent URI for this collectionhttp://hdl.handle.net/1903/3

New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a given thesis/dissertation in DRUM.

More information is available at Theses and Dissertations at University of Maryland Libraries.

Browse

Subject: search.filters.subject.convex optimization

Search Results

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    CUR Matrix Approximation Through Convex Optimization
    (2024) Linehan, Kathryn; Balan, Radu V; Applied Mathematics and Scientific Computation; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    In this dissertation we present work on the CUR matrix approximation. Specifically, we present 1) an approximation of the proximal operator of the L-infinity norm using a neural network, 2) a novel deterministic CUR formulation and algorithm, and 3) a novel application of CUR as a feature selection method to determine discriminant proteins when clustering protein expression data in a self-organizing map (SOM). The proximal operator of the L-infinity norm arises in our CUR algorithm. Since the computation of the proximal operator of the L-infinity norm requires a sort of the input data (or at least a partial sort similar to quicksort), we present a neural network to approximate the proximal operator. A novel aspect of the network is that it is able to accept vectors of varying lengths due to a feature selection process that uses moments of the input data. We present results on the accuracy of the approximation, feature importance, and computational efficiency of the approach, and present an algorithm to calculate the proximal operator of the L-infinity norm exactly, relate it to the Moreau decomposition, and compare its computational efficiency to that of the approximation. Next, we present a novel deterministic CUR formulation that uses convex optimization to form the matrices C and R, and a corresponding algorithm that uses bisection to ensure that the user selected number of columns appear in C and the user selected number of rows appear in R. We implement the algorithm using the surrogate functional technique of Daubechies et al. [Communications on Pure and Applied Mathematics, 57.11 (2004)] and extend the theory of this approach to apply to our CUR formulation. Numerical results are presented that demonstrate the effectiveness of our CUR algorithm as compared to the singular value decomposition (SVD) and other CUR algorithms. Last, we use our CUR approximation as a feature selection method in the application by Higuera et al. [PLOS ONE, 10(6) (2015)] to determine discriminant proteins when clustering protein expression data in an SOM. This is a novel application of CUR and to the best of our knowledge, this is the first use of CUR on protein expression data. We compare the performance of our CUR algorithm to other CUR algorithms and the Wilcoxon rank-sum test (the original feature selection method in the work).
  • Thumbnail Image
    Item
    Jointly Optimal Placement and Power Allocation of Wireless Networks
    (2008-04-28) firouzabadi, sina; Martins, Nuno C; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    In this thesis, we investigate the optimal design of wireless networks. We consider wireless networks that have fixed and movable nodes, and we assume that all nodes feature adjustable transmission power. Hence, we aim at maximizing network centric objectives, by optimizing over admissible choices of the positions of the movable nodes as well as the transmission power at all the nodes. We adopt exponential path loss, which is a realistic assumption in urban and sub sea environments, and we propose ways of using this assumption to obtain a tractable optimization problem. Our formulation allows for the optimization of typical network centric objectives, such as power and throughput. It also allows signal-to-interference based constraints, such as rate-regions and outage probabilities, under the high signal to interference regime. We show that our optimization paradigm is convex and that it can be solved up to an arbitrary degree of accuracy via geometric programming techniques. By using a primal-dual decomposition, we also provide a case-study that illustrates how certain instances of our optimization paradigm can be solved via distributed iterative algorithms. We show that such a solution method also leads to a convenient layering in the primal step, whereby the power allocation and the node placement become two independent sub-problems.