-
Neighborhood Watch: Representation Learning with Local-Margin Triplet Loss and Sampling Strategy for K-Nearest-Neighbor Image Classification
Authors:
Phawis Thammasorn,
Daniel Hippe,
Wanpracha Chaovalitwongse,
Matthew Spraker,
Landon Wootton,
Matthew Nyflot,
Stephanie Combs,
Jan Peeken,
Eric Ford
Abstract:
Deep representation learning using triplet network for classification suffers from a lack of theoretical foundation and difficulty in tuning both the network and classifiers for performance. To address the problem, local-margin triplet loss along with local positive and negative mining strategy is proposed with theory on how the strategy integrate nearest-neighbor hyper-parameter with triplet lear…
▽ More
Deep representation learning using triplet network for classification suffers from a lack of theoretical foundation and difficulty in tuning both the network and classifiers for performance. To address the problem, local-margin triplet loss along with local positive and negative mining strategy is proposed with theory on how the strategy integrate nearest-neighbor hyper-parameter with triplet learning to increase subsequent classification performance. Results in experiments with 2 public datasets, MNIST and Cifar-10, and 2 small medical image datasets demonstrate that proposed strategy outperforms end-to-end softmax and typical triplet loss in settings without data augmentation while maintaining utility of transferable feature for related tasks. The method serves as a good performance baseline where end-to-end methods encounter difficulties such as small sample data with limited allowable data augmentation.
△ Less
Submitted 28 October, 2019;
originally announced November 2019.
-
An Implicit Cover Problem in Wild Population Study
Authors:
Mary V. Ashley,
Tanya Y. Berger-Wolf,
Wanpracha Chaovalitwongse,
Bhaskar DasGupta,
Ashfaq Khokhar,
Saad Sheikh
Abstract:
In an implicit combinatorial optimization problem, the constraints are not enumerated explicitly but rather stated implicitly through equations, other constraints or auxiliary algorithms. An important subclass of such problems is the implicit set cover (or, equivalently, hitting set) problem in which the sets are not given explicitly but rather defined implicitly For example, the well-known minimu…
▽ More
In an implicit combinatorial optimization problem, the constraints are not enumerated explicitly but rather stated implicitly through equations, other constraints or auxiliary algorithms. An important subclass of such problems is the implicit set cover (or, equivalently, hitting set) problem in which the sets are not given explicitly but rather defined implicitly For example, the well-known minimum feedback arc set problem is such a problem. In this paper, we consider such a cover problem that arises in the study of wild populations in biology in which the sets are defined implicitly via the Mendelian constraints and prove approximability results for this problem.
△ Less
Submitted 26 February, 2011;
originally announced February 2011.
-
On Approximating Four Covering and Packing Problems
Authors:
Mary Ashley,
Tanya Berger-Wolf,
Piotr Berman,
Wanpracha Chaovalitwongse,
Bhaskar DasGupta,
Ming-Yang Kao
Abstract:
In this paper, we consider approximability issues of the following four problems: triangle packing, full sibling reconstruction, maximum profit coverage and 2-coverage. All of them are generalized or specialized versions of set-cover and have applications in biology ranging from full-sibling reconstructions in wild populations to biomolecular clusterings; however, as this paper shows, their approx…
▽ More
In this paper, we consider approximability issues of the following four problems: triangle packing, full sibling reconstruction, maximum profit coverage and 2-coverage. All of them are generalized or specialized versions of set-cover and have applications in biology ranging from full-sibling reconstructions in wild populations to biomolecular clusterings; however, as this paper shows, their approximability properties differ considerably. Our inapproximability constant for the triangle packing problem improves upon the previous results; this is done by directly transforming the inapproximability gap of Haastad for the problem of maximizing the number of satisfied equations for a set of equations over GF(2) and is interesting in its own right. Our approximability results on the full siblings reconstruction problems answers questions originally posed by Berger-Wolf et al. and our results on the maximum profit coverage problem provides almost matching upper and lower bounds on the approximation ratio, answering a question posed by Hassin and Or.
△ Less
Submitted 4 February, 2011;
originally announced February 2011.
