8 results on '"Equivalent Representations"'
Search Results
2. Representation of spatial transformations in deep neural networks
- Author
-
Lenc, Karel, Zisserman, Andrew, Vedaldi, Andrea, and Lepetit, Vincent
- Subjects
006.3 ,Computer vision ,Geometric Equivariance ,Image Representations ,Convolutional Neural Networks ,Equivalent Representations ,Deep Learning - Abstract
This thesis addresses the problem of investigating the properties and abilities of a variety of computer vision representations with respect to spatial geometric transformations. Our approach is to employ machine learning methods for finding the behaviour of existing image representations empirically and to apply deep learning to new computer vision tasks where the underlying spatial information is of importance. The results help to further the understanding of modern computer vision representations, such as convolutional neural networks (CNNs) in image classification and object detection and to enable their application to new domains such as local feature detection. Because our theoretical understanding of CNNs remains limited, we investigate two key mathematical properties of representations: equivariance (how transformations of the input image are encoded) and equivalence (how two representations, for example two different parameterizations, layers or architectures share the same visual information). A number of methods to establish these properties empirically are proposed. These methods reveal interesting aspects of their structure, including clarifying at which layers in a CNN geometric invariances are achieved and how various CNN architectures differ. We identify several predictors of geometric and architectural compatibility. Direct applications to structured-output regression are demonstrated as well. Local covariant feature detection has been difficult to approach with machine learning techniques. We propose the first fully general formulation for learning local covariant feature detectors which casts detection as a regression problem, enabling the use of powerful regressors such as deep neural networks. The derived covariance constraint can be used to automatically learn which visual structures provide stable anchors for local feature detection. We support these ideas theoretically, and show that existing detectors can be derived in this framework. Additionally, in cooperation with Imperial College London, we introduce a novel large-scale dataset for evaluation of local detectors and descriptors. It is suitable for training and testing modern local features, together with strictly defined evaluation protocols for descriptors in several tasks such as matching, retrieval and verification. The importance of pixel-wise image geometry for object detection is unknown as the best results used to be obtained with combination of CNNs with cues from image segmentation. We propose a detector which uses constant region proposals and, while it approximates objects poorly, we show that a bounding box regressor using intermediate convolutional features can recover sufficiently accurate bounding boxes, demonstrating that the required geometric information is contained in the CNN itself. Combined with other improvements, we obtain an excellent and fast detector that processes an image only with the CNN.
- Published
- 2017
3. The quadratic cycle cover problem: special cases and efficient bounds.
- Author
-
de Meijer, Frank and Sotirov, Renata
- Abstract
The quadratic cycle cover problem is the problem of finding a set of node-disjoint cycles visiting all the nodes such that the total sum of interaction costs between consecutive arcs is minimized. In this paper we study the linearization problem for the quadratic cycle cover problem and related lower bounds. In particular, we derive various sufficient conditions for the quadratic cost matrix to be linearizable, and use these conditions to compute bounds. We also show how to use a sufficient condition for linearizability within an iterative bounding procedure. In each step, our algorithm computes the best equivalent representation of the quadratic cost matrix and its optimal linearizable matrix with respect to the given sufficient condition for linearizability. Further, we show that the classical Gilmore–Lawler type bound belongs to the family of linearization based bounds, and therefore apply the above mentioned iterative reformulation technique. We also prove that the linearization vectors resulting from this iterative approach satisfy the constant value property. The best among here introduced bounds outperform existing lower bounds when taking both quality and efficiency into account. [ABSTRACT FROM AUTHOR]
- Published
- 2020
- Full Text
- View/download PDF
4. Representation, optimization and generation of fuzzy measures.
- Author
-
Beliakov, Gleb, Wu, Jian-Zhang, and Ding, Weiping
- Subjects
- *
FUZZY measure theory , *FUZZY integrals , *RANDOM measures , *FUZZY sets , *COMPUTATIONAL complexity , *PRICES , *DECISION making - Abstract
We review recent literature on three aspects of fuzzy measures: their representations, learning optimal fuzzy measures and random generation of various types of fuzzy measures. These three aspects are interdependent: methods of learning fuzzy measures depend on their representation, and may also include random generation as one of the steps, on the other hand different representations also affect generation methods, while random generation plays an important role in simulation studies for post-hoc analysis of sets of measures learned from data and problem-specific constraints. Explicit modelling of interactions between the decision variables is a distinctive feature of integrals based on fuzzy measures, but its price is high computational complexity. To extend their range of applicability efficient representations and computational techniques are required. All three mentioned aspects provide mathematical and computational tools for novel applications of fuzzy measures and integrals in decision making and information fusion, allow scaling up significantly the domain of applicability and reduce their complexity. • Recent developments on three aspects of fuzzy measures surveyed. • New efficient methods of learning fuzzy measures from data reviewed. • New methods of random generation fuzzy measures compared. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF
5. Understanding Image Representations by Measuring Their Equivariance and Equivalence.
- Author
-
Lenc, Karel and Vedaldi, Andrea
- Subjects
- *
IMAGE representation , *GEOMETRIC analysis , *MATHEMATICAL equivalence , *NEURAL computers , *COMPUTER architecture - Abstract
Despite the importance of image representations such as histograms of oriented gradients and deep Convolutional Neural Networks (CNN), our theoretical understanding of them remains limited. Aimed at filling this gap, we investigate two key mathematical properties of representations: equivariance and equivalence. Equivariance studies how transformations of the input image are encoded by the representation, invariance being a special case where a transformation has no effect. Equivalence studies whether two representations, for example two different parameterizations of a CNN, two different layers, or two different CNN architectures, share the same visual information or not. A number of methods to establish these properties empirically are proposed, including introducing transformation and stitching layers in CNNs. These methods are then applied to popular representations to reveal insightful aspects of their structure, including clarifying at which layers in a CNN certain geometric invariances are achieved and how various CNN architectures differ. We identify several predictors of geometric and architectural compatibility, including the spatial resolution of the representation and the complexity and depth of the models. While the focus of the paper is theoretical, direct applications to structured-output regression are demonstrated too. [ABSTRACT FROM AUTHOR]
- Published
- 2019
- Full Text
- View/download PDF
6. A general representation theory for constructing groups of permutation polynomials.
- Author
-
Castillo, Chris and Coulter, Robert S.
- Subjects
- *
REPRESENTATION theory , *PERMUTATIONS , *POLYNOMIALS , *NONABELIAN groups , *GEOMETRICAL constructions , *FINITE fields - Abstract
Using the left regular action of a group on itself, we develop a general representation theory for constructing groups of permutation polynomials. As an application of the method, we compute polynomial representations of several abelian and nonabelian groups, and we determine the equivalence classes of the groups of polynomials we construct. In particular, when the size of the group is equal to the size of the field in which the group is represented, all non-identity representation polynomials are necessarily fixed-point free permutation polynomials. [ABSTRACT FROM AUTHOR]
- Published
- 2015
- Full Text
- View/download PDF
7. On hereditary irreducible unimonomial representations of cyclic p-groups over local rings of characteristic
- Subjects
hereditary irreducible representation ,monomial matrix ,GAP ,еквiвалентнi зображення ,унiмономiальне зображення ,unimonomial representation ,спадково незвiдне зображення ,мономiальна матриця ,equivalent representations - Abstract
The task of the description up to equivalency of the matrix representations of finite pgroups of order greater then p aver a commutative local ring of characteristics of ps (s > 0) that is not a field contains the classical unsolved problem of pair of matrices over a field. Therefore, consideration of partial cases and the study of special representation matrix representations is importent. Let K be a commutative ring with a identity, G is a finitely generated group with some fixed system of generator elements a1,...,ar. Every matrix representation of the group G over the ring K equivalent to Γ : ai → E+Mi (i = 1,...,r), where E be an identity n×n-matrix, Mi be a monomial n×n-matrix (i = 1,...,r), we call unimonomial. An reducible unimonomial representation over the ring K, in a reducible form of which on diagonal blocks at least one unimonomial representation is formed, we call hereditary reducible over the ring K. A series of unimonomial representations of a finite cyclic p-group H = haiover commutative local principle ideal ring K of characteristic p with nilpotent Jacobson radical of the degree l (1 < l < ∞) of the form a → E + 0 ... 0 1 1 ... 0 0 . . . ... . . . . . . 0 ... 1 0!· diag[ε1ts1,...,εntsn], where si ≥ 0, εi be elements from K∗ (i = 1,...,n), t be an generator element of Jacobson radical ring K. It is making up clear the criterion, when the map of the given form sets the representation of the group H (P|H|−1 j=0 si+j ≥ l (i = 1,...,n), herethe indexes are considered by the module n). It have been found the sufficient condition of hereditary irreducibility of the constructed representations ((Pn i=1 si,n) = 1,tPn i=1 si 6= 0). In addition, we can obtain the hereditary reducibility of the constructed representations in the case when (Pn i=1 si,n) > 1,Qn i=1 εi = 1). Based on the researches of Bondarenko V. M., Bortosh M. Yu. of similarity of the monomial matrices it is making up clear the criterion the equivalence of the constructed representations (the corresponding sequences (s1,...,sn) are cyclically equivalent a relevant productsQn i=1 εi are equal modulo Ann (ts), where s is the largest member of the weight sequence (s1,...,sn)). In the case of the finiteness of the ring K by computation in the GAP system it have been found the number of all, up to equivalence, constructed unimonomial hereditary irreducible matrix representations of a cyclic nontrivial p-group depending on the number of elements of the residue class field of the ring K., Задача про опис з точнiстю до еквiвалентностi матричних зображень скiнченних pгруппорядкувище p надкомутативнимлокальнимкiльцемхарактеристики ps (s > 0), що не є полем, мiстить у собi класичну нерозв’язну задачу про пару матриць над полем. Тому актуальним є розгляд частинних випадкiв i вивчення матричних зображень спецiальноговигляду.Нехай K —комутативнекiльцезодиницею, G —скiнченнопородженагрупаздеякоюфiксованоюсистемоютвiрнихелементiв a1,...,ar.Всякематричне зображення групи G над кiльцем K еквiвалентне до Γ : ai → E+Mi (i = 1,...,r), де E — одинична матриця порядку n, Mi — мономiальна матриця порядку n (i = 1,...,r), назвемо унiмономiальним. Звiдне унiмономiальне зображення над кiльцем K, в зведеному виглядi якого на дiагональних блоках утворюється хоч одне унiмономiальне зображення, назвемо спадково звiдним над кiльцем K. Побудовано серiю унiмономiальних зображень скiнченної циклiчної p-група H = hai над комутативним локальним кiльцем K головних iдеалiв характеристики p з нiльпотентностим радикалом Джекобсона ступуня l (1 < l < ∞), вигляду a → E + 0 ... 0 1 1 ... 0 0 . . . ... . . . . . . 0 ... 1 0!·diag[ε1ts1,...,εntsn], де si ≥ 0, εi — елементи iз K∗ (i = 1,...,n), t — твiрний елемент радикалу Джекобсона кiльця K. З’ясовано критерiй, коли вiдображення заданого вигляду задає зображення групи H (P|H|−1 j=0 si+j ≥ l (i = 1,...,n), тут iндекси розглядаються за модулем n). З’ясованодостатнюумовуспадковоїнезвiдностiпобудованихзображень((Pn i=1 si,n) = 1,P n i=1 si < l). Крiм того, можна отримати спадкову звiднiсть побудованих зображень у випадку, коли (Pn i=1 si,n) > 1,Qn i=1 εi = 1). На основi дослiджень Бондаренка В. М., Бортош М. Ю. подiбностi мономiальних матриць з’ясовано критерiй еквiвалентностi побудованих зображень (вiдповiднi послiдовностi (s1,...,sn) циклiчно еквiвалентнi а вiдповiднi добуткиQn i=1 εi рiвнi за модулем Ann(ts), де s — найбiльший член вагової послiдовностi (s1,...,sn)). У випадку скiнченностi кiльця K засобами обчислень в системiGAPпорахованочисловсiх,зточнiстюдоеквiвалентностi,побудованихунiмономiальних спадково незвiднi матричних зображень циклiчної нетривiальної p-групи залежно вiд числа елементiв поля лишкiв кiльця K.
- Published
- 2019
- Full Text
- View/download PDF
8. The Quadratic Cycle Cover Problem: special cases and efficient bounds
- Author
-
Frank de Meijer, Renata Sotirov, Econometrics and Operations Research, Research Group: Operations Research, and Center Ph. D. Students
- Subjects
Control and Optimization ,Linearizability ,0211 other engineering and technologies ,0102 computer and information sciences ,02 engineering and technology ,linearization problem ,01 natural sciences ,Matrix (mathematics) ,Quadratic equation ,Linearization ,Bounding overwatch ,Quadratic cycle cover problem ,FOS: Mathematics ,Discrete Mathematics and Combinatorics ,Applied mathematics ,Representation (mathematics) ,Mathematics - Optimization and Control ,Mathematics ,021103 operations research ,Applied Mathematics ,Computer Science Applications ,Computational Theory and Mathematics ,Optimization and Control (math.OC) ,010201 computation theory & mathematics ,Theory of computation ,Gilmore-Lawler bound ,Constant (mathematics) ,equivalent representations - Abstract
The quadratic cycle cover problem is the problem of finding a set of node-disjoint cycles visiting all the nodes such that the total sum of interaction costs between consecutive arcs is minimized. In this paper we study the linearization problem for the quadratic cycle cover problem and related lower bounds. In particular, we derive various sufficient conditions for the quadratic cost matrix to be linearizable, and use these conditions to compute bounds. We also show how to use a sufficient condition for linearizability within an iterative bounding procedure. In each step, our algorithm computes the best equivalent representation of the quadratic cost matrix and its optimal linearizable matrix with respect to the given sufficient condition for linearizability. Further, we show that the classical Gilmore–Lawler type bound belongs to the family of linearization based bounds, and therefore apply the above mentioned iterative reformulation technique. We also prove that the linearization vectors resulting from this iterative approach satisfy the constant value property. The best among here introduced bounds outperform existing lower bounds when taking both quality and efficiency into account.
- Published
- 2019
Catalog
Discovery Service for Jio Institute Digital Library
For full access to our library's resources, please sign in.