-
Revealing data leakage in protein interaction benchmarks
Authors:
Anton Bushuiev,
Roman Bushuiev,
Jiri Sedlar,
Tomas Pluskal,
Jiri Damborsky,
Stanislav Mazurenko,
Josef Sivic
Abstract:
In recent years, there has been remarkable progress in machine learning for protein-protein interactions. However, prior work has predominantly focused on improving learning algorithms, with less attention paid to evaluation strategies and data preparation. Here, we demonstrate that further development of machine learning methods may be hindered by the quality of existing train-test splits. Specif…
▽ More
In recent years, there has been remarkable progress in machine learning for protein-protein interactions. However, prior work has predominantly focused on improving learning algorithms, with less attention paid to evaluation strategies and data preparation. Here, we demonstrate that further development of machine learning methods may be hindered by the quality of existing train-test splits. Specifically, we find that commonly used splitting strategies for protein complexes, based on protein sequence or metadata similarity, introduce major data leakage. This may result in overoptimistic evaluation of generalization, as well as unfair benchmarking of the models, biased towards assessing their overfitting capacity rather than practical utility. To overcome the data leakage, we recommend constructing data splits based on 3D structural similarity of protein-protein interfaces and suggest corresponding algorithms. We believe that addressing the data leakage problem is critical for further progress in this research area.
△ Less
Submitted 16 April, 2024;
originally announced April 2024.
-
Domination number of modular product graphs
Authors:
Sergio Bermudo,
Iztok Peterin,
Jelena Sedlar,
Riste Škrekovski
Abstract:
The modular product $G\diamond H$ of graphs $G$ and $H$ is a graph on vertex set $V(G)\times V(H)$. Two vertices $(g,h)$ and $(g^{\prime},h^{\prime})$ of $G\diamond H$ are adjacent if $g=g^{\prime}$ and $hh^{\prime}\in E(H)$, or $gg^{\prime}\in E(G)$ and $h=h^{\prime}$, or $gg^{\prime}\in E(G)$ and $hh^{\prime}\in E(H)$, or (for $g\neq g^{\prime}$ and $h\neq h^{\prime}$) $gg^{\prime}\notin E(G)$ a…
▽ More
The modular product $G\diamond H$ of graphs $G$ and $H$ is a graph on vertex set $V(G)\times V(H)$. Two vertices $(g,h)$ and $(g^{\prime},h^{\prime})$ of $G\diamond H$ are adjacent if $g=g^{\prime}$ and $hh^{\prime}\in E(H)$, or $gg^{\prime}\in E(G)$ and $h=h^{\prime}$, or $gg^{\prime}\in E(G)$ and $hh^{\prime}\in E(H)$, or (for $g\neq g^{\prime}$ and $h\neq h^{\prime}$) $gg^{\prime}\notin E(G)$ and $hh^{\prime}\notin E(H)$. A set $D\subseteq V(G)$ is a dominating set of $G$ if every vertex outside of $D$ contains a neighbor in $D$. A set $D\subseteq V(G)$ is a total dominating set of $G$ if every vertex of $G$ contains a neighbor in $D$. The domination number $γ(G)$ (resp. total domination number $γ_{t}(G)$) of $G$ is the minimum cardinality of a dominating set (resp. total dominating set) of $G$. In this work we give several upper and lower bounds for $γ(G\diamond H)$ in terms of $γ(G),$ $γ(H)$, $γ_{t}(\overline{G})$ and $γ_{t}(\overline{H})$, where $\overline{G}$ is the complement graph of $G$. Further, we fully describe graphs where $γ(G\diamond H)=k$ for $k\in\{1,2,3\}$. Several conditions on $G$ and $H$ under which $γ(G\diamond H)$ is at most $4$ and $5$ are also given. A new type of simultaneous domination $\barγ(G)$, defined as the smallest number of vertices that dominates $G$ and totally dominates the complement of $G,$ emerged as useful and we believe it could be of independent interest. We conclude the paper by proposing few directions for possible further research.
△ Less
Submitted 3 April, 2024;
originally announced April 2024.
-
Proper Z4 x Z2-colorings: structural characterization with application to some snarks
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
A proper abelian coloring of a cubic graph G by a finite abelian group A is any proper edge-coloring of G by the non-zero elements of A such that the sum of the colors of the three edges incident to any vertex v of G equals zero. It is known that cyclic groups of order smaller than 10 do not color all bridgeless cubic graphs, and that all abelian groups of order at least 12 do. This leaves the que…
▽ More
A proper abelian coloring of a cubic graph G by a finite abelian group A is any proper edge-coloring of G by the non-zero elements of A such that the sum of the colors of the three edges incident to any vertex v of G equals zero. It is known that cyclic groups of order smaller than 10 do not color all bridgeless cubic graphs, and that all abelian groups of order at least 12 do. This leaves the question open for the four so called exceptional groups Z4 x Z2, Z3 x Z3, Z10 and Z11 for snarks. It is conjectured in literature that every cubic graph has a proper abelian coloring by each exceptional group and it is further known that the existence of a proper Z4 x Z2-coloring of G implies the existence of a proper coloring of G by all the remaining exceptional groups. In this paper, we give a characterization of a proper Z4 x Z2-coloring in terms of the existence of a matching M in a 2-factor F of G with particular properties. Moreover, in order to modify an arbitrary matching M so that it meets the requirements of the characterization, we first introduce an incidence structure of the cycles of F in relation to the cycles of G - M. Further, we provide a sufficient condition under which M can be modified into a desired matching in terms of particular properties of the introduced incidence structure. We conclude the paper by applying the results to some oddness two snarks, in particular to permutation snarks. We believe that the approach of this paper with some additional refinements extends to larger classes of snarks, if not to all in general.
△ Less
Submitted 8 February, 2024;
originally announced February 2024.
-
Normal 5-edge coloring of some more snarks superpositioned by the Petersen graph
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
A normal 5-edge-coloring of a cubic graph is a coloring such that for every edge the number of distinct colors incident to its end-vertices is 3 or 5 (and not 4). The well known Petersen Coloring Conjecture is equivalent to the statement that every bridgeless cubic graph has a normal 5-edge-coloring. All 3-edge-colorings of a cubic graph are obviously normal, so in order to establish the conjectur…
▽ More
A normal 5-edge-coloring of a cubic graph is a coloring such that for every edge the number of distinct colors incident to its end-vertices is 3 or 5 (and not 4). The well known Petersen Coloring Conjecture is equivalent to the statement that every bridgeless cubic graph has a normal 5-edge-coloring. All 3-edge-colorings of a cubic graph are obviously normal, so in order to establish the conjecture it is sufficient to consider only snarks. In our previous paper [J. Sedlar, R. Škrekovski, Normal 5-edge-coloring of some snarks superpositioned by the Petersen graph, Applied Mathematics and Computation 467 (2024) 128493], we considered superpositions of any snark G along a cycle C by two simple supervertices and by the superedge obtained from the Petersen graph, but only for some of the possible ways of connecting supervertices and superedges. The present paper is a continuation of that paper, herein we consider superpositions by the Petersen graph for all the remaining connections and establish that for all of them the Petersen Coloring Conjecture holds.
△ Less
Submitted 15 December, 2023; v1 submitted 14 December, 2023;
originally announced December 2023.
-
Learning to design protein-protein interactions with enhanced generalization
Authors:
Anton Bushuiev,
Roman Bushuiev,
Petr Kouba,
Anatolii Filkin,
Marketa Gabrielova,
Michal Gabriel,
Jiri Sedlar,
Tomas Pluskal,
Jiri Damborsky,
Stanislav Mazurenko,
Josef Sivic
Abstract:
Discovering mutations enhancing protein-protein interactions (PPIs) is critical for advancing biomedical research and develo** improved therapeutics. While machine learning approaches have substantially advanced the field, they often struggle to generalize beyond training data in practical scenarios. The contributions of this work are three-fold. First, we construct PPIRef, the largest and non-r…
▽ More
Discovering mutations enhancing protein-protein interactions (PPIs) is critical for advancing biomedical research and develo** improved therapeutics. While machine learning approaches have substantially advanced the field, they often struggle to generalize beyond training data in practical scenarios. The contributions of this work are three-fold. First, we construct PPIRef, the largest and non-redundant dataset of 3D protein-protein interactions, enabling effective large-scale learning. Second, we leverage the PPIRef dataset to pre-train PPIformer, a new SE(3)-equivariant model generalizing across diverse protein-binder variants. We fine-tune PPIformer to predict effects of mutations on protein-protein interactions via a thermodynamically motivated adjustment of the pre-training loss function. Finally, we demonstrate the enhanced generalization of our new PPIformer approach by outperforming other state-of-the-art methods on new, non-leaking splits of standard labeled PPI mutational data and independent case studies optimizing a human antibody against SARS-CoV-2 and increasing the thrombolytic activity of staphylokinase.
△ Less
Submitted 16 March, 2024; v1 submitted 27 October, 2023;
originally announced October 2023.
-
Resolving vertices of graphs with differences
Authors:
Iztok Peterina,
Jelena Sedlar,
Riste Škrekovski,
Ismael G. Yero
Abstract:
The classical (vertex) metric dimension of a graph G is defined as the cardinality of a smallest set S in V (G) such that any two vertices x and y from G have different distances to least one vertex from S: The k-metric dimension is a generalization of that notion where it is required that any pair of vertices has different distances to at least k vertices from S: In this paper, we introduce the w…
▽ More
The classical (vertex) metric dimension of a graph G is defined as the cardinality of a smallest set S in V (G) such that any two vertices x and y from G have different distances to least one vertex from S: The k-metric dimension is a generalization of that notion where it is required that any pair of vertices has different distances to at least k vertices from S: In this paper, we introduce the weak k-metric dimension of a graph G; which is defined as the cardinality of a smallest set of vertices S such that the sum of the distance differences from any pair of vertices to all vertices of S is at least k: This dimension is "stronger" than the classical metric dimension, yet "weaker" than k-metric dimension, and it can be formulated as an ILP problem. The maximum k for which the weak k-metric dimension is defined is denoted by kappa(G). We first prove several properties of the weak k-metric dimension regarding the presence of true or false twin vertices in a graph. Using those properties, the kappa(G) is found for some basic graph classes, such as paths, stars, cycles, and complete (bipartite) graphs. We also find kappa(G) for trees and grid graphs using the observation that the distance difference increases by the increase of the cardinality of a set S. For all these graph classes we further establish the exact value of the weak k-metric dimension for all k <= kappa(G).
△ Less
Submitted 2 September, 2023;
originally announced September 2023.
-
Normal 5-edge-coloring of some snarks superpositioned by Flower snarks
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
An edge e is normal in a proper edge-coloring of a cubic graph G if the number of distinct colors on four edges incident to e is 2 or 4: A normal edge-coloring of G is a proper edge-coloring in which every edge of G is normal. The Petersen Coloring Conjecture is equivalent to stating that every bridgeless cubic graph has a normal 5-edge-coloring. Since every 3-edge-coloring of a cubic graph is tri…
▽ More
An edge e is normal in a proper edge-coloring of a cubic graph G if the number of distinct colors on four edges incident to e is 2 or 4: A normal edge-coloring of G is a proper edge-coloring in which every edge of G is normal. The Petersen Coloring Conjecture is equivalent to stating that every bridgeless cubic graph has a normal 5-edge-coloring. Since every 3-edge-coloring of a cubic graph is trivially normal, it is suficient to consider only snarks to establish the conjecture. In this paper, we consider a class of superpositioned snarks obtained by choosing a cycle C in a snark G and superpositioning vertices of C by one of two simple supervertices and edges of C by superedges Hx;y, where H is any snark and x; y any pair of nonadjacent vertices of H: For such superpositioned snarks, two suficient conditions are given for the existence of a normal 5-edge-coloring. The first condition yields a normal 5-edge-coloring for all hypohamiltonian snarks used as superedges, but only for some of the possible ways of connecting them. In particular, since the Flower snarks are hypohamiltonian, this consequently yields a normal 5-edge-coloring for many snarks superpositioned by the Flower snarks. The second sufficient condition is more demanding, but its application yields a normal 5-edge-colorings for all superpositions by the Flower snarks. The same class of snarks is considered in [S. Liu, R.-X. Hao, C.-Q. Zhang, Berge{Fulkerson coloring for some families of superposition snarks, Eur. J. Comb. 96 (2021) 103344] for the Berge-Fulkerson conjecture. Since we established that this class has a Petersen coloring, this immediately yields the result of the above mentioned paper.
△ Less
Submitted 23 June, 2023;
originally announced June 2023.
-
Normal 5-edge-coloring of some snarks superpositioned by the Petersen graph
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
In a (proper) edge-coloring of a bridgeless cubic graph G an edge e is rich (resp. poor) if the number of colors of all edges incident to end-vertices of e is 5 (resp. 3). An edge-coloring of G is is normal if every edge of G is either rich or poor. In this paper we consider snarks ~G obtained by a simple superposition of edges and vertices of a cycle C in a snark G: For an even cycle C we show th…
▽ More
In a (proper) edge-coloring of a bridgeless cubic graph G an edge e is rich (resp. poor) if the number of colors of all edges incident to end-vertices of e is 5 (resp. 3). An edge-coloring of G is is normal if every edge of G is either rich or poor. In this paper we consider snarks ~G obtained by a simple superposition of edges and vertices of a cycle C in a snark G: For an even cycle C we show that a normal coloring of G can be extended to a normal coloring of ~G without changing colors of edges outside C in G: An interesting remark is that this is in general impossible for odd cycles, since the normal coloring of a Petersen graph P10 cannot be extended to a superposition of P10 on a 5-cycle without changing colors outside the 5-cycle. On the other hand, as our colorings of the superpositioned snarks introduce 18 or more poor edges, we are inclined to believe that every bridgeless cubic graph istinct from P10 has a normal coloring with at least one poor edge and possibly with at least 6 if we also exclude the Petersen graph with one vertex truncated.
△ Less
Submitted 10 May, 2023;
originally announced May 2023.
-
Imitrob: Imitation Learning Dataset for Training and Evaluating 6D Object Pose Estimators
Authors:
Jiri Sedlar,
Karla Stepanova,
Radoslav Skoviera,
Jan K. Behrens,
Matus Tuna,
Gabriela Sejnova,
Josef Sivic,
Robert Babuska
Abstract:
This paper introduces a dataset for training and evaluating methods for 6D pose estimation of hand-held tools in task demonstrations captured by a standard RGB camera. Despite the significant progress of 6D pose estimation methods, their performance is usually limited for heavily occluded objects, which is a common case in imitation learning, where the object is typically partially occluded by the…
▽ More
This paper introduces a dataset for training and evaluating methods for 6D pose estimation of hand-held tools in task demonstrations captured by a standard RGB camera. Despite the significant progress of 6D pose estimation methods, their performance is usually limited for heavily occluded objects, which is a common case in imitation learning, where the object is typically partially occluded by the manipulating hand. Currently, there is a lack of datasets that would enable the development of robust 6D pose estimation methods for these conditions. To overcome this problem, we collect a new dataset (Imitrob) aimed at 6D pose estimation in imitation learning and other applications where a human holds a tool and performs a task. The dataset contains image sequences of nine different tools and twelve manipulation tasks with two camera viewpoints, four human subjects, and left/right hand. Each image is accompanied by an accurate ground truth measurement of the 6D object pose obtained by the HTC Vive motion tracking device. The use of the dataset is demonstrated by training and evaluating a recent 6D object pose estimation method (DOPE) in various setups.
△ Less
Submitted 5 April, 2023; v1 submitted 16 September, 2022;
originally announced September 2022.
-
Local Irregularity Conjecture vs. cacti
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
A graph is locally irregular if the degrees of the end-vertices of every edge are distinct. An edge coloring of a graph G is locally irregular if every color induces a locally irregular subgraph of G. A colorable graph G is any graph which admits a locally irregular edge coloring. The locally irregular chromatic index X'irr(G) of a colorable graph G is the smallest number of colors required by a l…
▽ More
A graph is locally irregular if the degrees of the end-vertices of every edge are distinct. An edge coloring of a graph G is locally irregular if every color induces a locally irregular subgraph of G. A colorable graph G is any graph which admits a locally irregular edge coloring. The locally irregular chromatic index X'irr(G) of a colorable graph G is the smallest number of colors required by a locally irregular edge coloring of G. The Local Irregularity Conjecture claims that all colorable graphs require at most 3 colors for locally irregular edge coloring. Recently, it has been observed that the conjecture does not hold for the bow-tie graph B [7]. Cacti are important class of graphs for this conjecture since B and all non-colorable graphs are cacti. In this paper we show that for every colorable cactus graph G != B it holds that X'irr(G) <= 3. This makes us to believe that B is the only colorable graph with X'irr(B) > 3, and consequently that B is the only counterexample to the Local Irregularity Conjecture.
△ Less
Submitted 20 July, 2022; v1 submitted 8 July, 2022;
originally announced July 2022.
-
A note on the locally irregular edge colorings of cacti
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
A graph is locally irregular if the degrees of the end-vertices of every edge are distinct. An edge coloring of a graph G is locally irregular if every color induces a locally irregular subgraph of G. A colorable graph G is any graph which admits a locally irregular edge coloring. The locally irregular chromatic index X'irr(G) of a colorable graph G is the smallest number of colors required by a l…
▽ More
A graph is locally irregular if the degrees of the end-vertices of every edge are distinct. An edge coloring of a graph G is locally irregular if every color induces a locally irregular subgraph of G. A colorable graph G is any graph which admits a locally irregular edge coloring. The locally irregular chromatic index X'irr(G) of a colorable graph G is the smallest number of colors required by a locally irregular edge coloring of G. The Local Irregularity Conjecture claims that all colorable graphs require at most 3 colors for a locally irregular edge coloring. Recently, it has been observed that the conjecture does not hold for the bow-tie graph B, since B is colorable and requires at least 4 colors for a locally irregular edge coloring. Since B is a cactus graph and all non-colorable graphs are also cacti, this seems to be a relevant class of graphs for the Local Irregularity Conjecture. In this paper we establish that X'irr(G)<= 4 for all colorable cactus graphs.
△ Less
Submitted 7 July, 2022;
originally announced July 2022.
-
Remarks on the vertex and the edge metric dimension of 2-connected graphs
Authors:
Martin Knor,
Jelena Sedlar,
Riste Škrekovski
Abstract:
The vertex (resp. edge) metric dimension of a graph G is the size of a smallest vertex set in G which distinguishes all pairs of vertices (resp. edges) in G and it is denoted by dim(G) (resp. edim(G)). The upper bounds dim(G) <= 2c(G) - 1 and edim(G) <= 2c(G)-1; where c(G) denotes the cyclomatic number of G, were established to hold for cacti without leaves distinct from cycles, and moreover all l…
▽ More
The vertex (resp. edge) metric dimension of a graph G is the size of a smallest vertex set in G which distinguishes all pairs of vertices (resp. edges) in G and it is denoted by dim(G) (resp. edim(G)). The upper bounds dim(G) <= 2c(G) - 1 and edim(G) <= 2c(G)-1; where c(G) denotes the cyclomatic number of G, were established to hold for cacti without leaves distinct from cycles, and moreover all leafless cacti which attain the bounds were characterized. It was further conjectured that the same bounds hold for general connected graphs without leaves and this conjecture was supported by showing that the problem reduces to 2-connected graphs. In this paper we focus on Theta graphs, as the most simple 2-connected graphs distinct from cycle, and show that the the upper bound 2c(G) - 1 holds for both metric dimensions of Theta graphs and we characterize all Theta graphs for which the bound is attained. We conclude by conjecturing that there are no other extremal graphs for the bound 2c(G) - 1 in the class of leafless graphs besides already known extremal cacti and extremal Theta graphs mentioned here.
△ Less
Submitted 14 March, 2022;
originally announced March 2022.
-
Remarks on the Local Irregularity Conjecture
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
A locally irregular graph is a graph in which the end-vertices of every edge have distinct degrees. A locally irregular edge coloring of a graph G is any edge coloring of G such that each of the colors induces a locally irregular subgraph of G. A graph G is colorable if it admits a locally irregular edge coloring. The locally irregular chromatic index of a colorable graph G, denoted by CHI'irr(G),…
▽ More
A locally irregular graph is a graph in which the end-vertices of every edge have distinct degrees. A locally irregular edge coloring of a graph G is any edge coloring of G such that each of the colors induces a locally irregular subgraph of G. A graph G is colorable if it admits a locally irregular edge coloring. The locally irregular chromatic index of a colorable graph G, denoted by CHI'irr(G), is the smallest number of colors used by a locally irregular edge coloring of G. The Local Irregularity Conjecture claims that all graphs, except odd length path, odd length cycle and a certain class of cacti, are colorable by 3 colors. As the conjecture is valid for graphs with large minimum degree and all non-colorable graphs are vertex disjoint cacti, we take direction to study rather sparse graphs. In this paper, we give a cactus graph B which contradicts this conjecture, i.e. CHI'irr(B) = 4. Nevertheless, we show that the conjecture holds for unicyclic graphs and cacti with vertex disjoint cycles.
△ Less
Submitted 15 November, 2021;
originally announced November 2021.
-
Estimating 3D Motion and Forces of Human-Object Interactions from Internet Videos
Authors:
Zongmian Li,
Jiri Sedlar,
Justin Carpentier,
Ivan Laptev,
Nicolas Mansard,
Josef Sivic
Abstract:
In this paper, we introduce a method to automatically reconstruct the 3D motion of a person interacting with an object from a single RGB video. Our method estimates the 3D poses of the person together with the object pose, the contact positions and the contact forces exerted on the human body. The main contributions of this work are three-fold. First, we introduce an approach to jointly estimate t…
▽ More
In this paper, we introduce a method to automatically reconstruct the 3D motion of a person interacting with an object from a single RGB video. Our method estimates the 3D poses of the person together with the object pose, the contact positions and the contact forces exerted on the human body. The main contributions of this work are three-fold. First, we introduce an approach to jointly estimate the motion and the actuation forces of the person on the manipulated object by modeling contacts and the dynamics of the interactions. This is cast as a large-scale trajectory optimization problem. Second, we develop a method to automatically recognize from the input video the 2D position and timing of contacts between the person and the object or the ground, thereby significantly simplifying the complexity of the optimization. Third, we validate our approach on a recent video+MoCap dataset capturing typical parkour actions, and demonstrate its performance on a new dataset of Internet videos showing people manipulating a variety of tools in unconstrained environments.
△ Less
Submitted 2 November, 2021;
originally announced November 2021.
-
Metric dimensions vs. cyclomatic number of graphs with minimum degree at least two
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
The vertex (resp. edge) metric dimension of a connected graph G; denoted by dim(G) (resp. edim(G)), is defined as the size of a smallest set S in V (G) which distinguishes all pairs of vertices (resp. edges) in G: Bounds dim(G) <= L(G)+2c(G) and edim(G) <= L(G) + 2c(G); where c(G) is the cyclomatic number in G and L(G) depends on the number of leaves in G, are known to hold for cacti and are conje…
▽ More
The vertex (resp. edge) metric dimension of a connected graph G; denoted by dim(G) (resp. edim(G)), is defined as the size of a smallest set S in V (G) which distinguishes all pairs of vertices (resp. edges) in G: Bounds dim(G) <= L(G)+2c(G) and edim(G) <= L(G) + 2c(G); where c(G) is the cyclomatic number in G and L(G) depends on the number of leaves in G, are known to hold for cacti and are conjectured to hold for general graphs. In leafless graphs it holds that L(G) = 0; so for such graphs the conjectured upper bound becomes 2c(G). In this paper, we show that the bound 2c(G) cannot be attained by leafless cacti, so the upper bound for such cacti decreases to 2c(G)-1, and we characterize all extremal leafless cacti for the decreased bound. We conjecture that the decreased bound holds for all leafless graphs, i.e. graphs with minimum degree at least two. We support this conjecture by showing that it holds for all graphs with minimum degree at least three and that it is suficient to show that it holds for all 2-connected graphs, and we also verify the conjecture for graphs of small order.
△ Less
Submitted 21 August, 2021;
originally announced August 2021.
-
Vertex and edge metric dimensions of cacti
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
In a graph G; a vertex (resp. an edge) metric generator is a set of vertices S such that any pair of vertices (resp. edges) from G is distinguished by at least one vertex from S: The cardinality of a smallest vertex (resp. edge) metric generator is the vertex (resp. edge) metric dimension of G: In [19] we determined the vertex (resp. edge) metric dimension of unicyclic graphs and that it takes its…
▽ More
In a graph G; a vertex (resp. an edge) metric generator is a set of vertices S such that any pair of vertices (resp. edges) from G is distinguished by at least one vertex from S: The cardinality of a smallest vertex (resp. edge) metric generator is the vertex (resp. edge) metric dimension of G: In [19] we determined the vertex (resp. edge) metric dimension of unicyclic graphs and that it takes its value from two consecutive integers. Therein, several cycle configurations were introduced and the vertex (resp. edge) metric dimension takes the greater of the two consecutive values only if any of these configurations is present in the graph. In this paper we extend the result to cactus graphs i.e. graphs in which all cycles are pairwise edge disjoint. We do so by defining a unicyclic subgraph of G for every cycle of G and applying the already introduced approach for unicyclic graphs which involves the configurations. The obtained results enable us to prove the cycle rank conjecture for cacti. They also yield a simple upper bound on metric dimensions of cactus graphs and we conclude the paper by conjecturing that the same upper bound holds in general.
△ Less
Submitted 3 July, 2021;
originally announced July 2021.
-
Learning to solve geometric construction problems from images
Authors:
J. Macke,
J. Sedlar,
M. Olsak,
J. Urban,
J. Sivic
Abstract:
We describe a purely image-based method for finding geometric constructions with a ruler and compass in the Euclidea geometric game. The method is based on adapting the Mask R-CNN state-of-the-art image processing neural architecture and adding a tree-based search procedure to it. In a supervised setting, the method learns to solve all 68 kinds of geometric construction problems from the first six…
▽ More
We describe a purely image-based method for finding geometric constructions with a ruler and compass in the Euclidea geometric game. The method is based on adapting the Mask R-CNN state-of-the-art image processing neural architecture and adding a tree-based search procedure to it. In a supervised setting, the method learns to solve all 68 kinds of geometric construction problems from the first six level packs of Euclidea with an average 92% accuracy. When evaluated on new kinds of problems, the method can solve 31 of the 68 kinds of Euclidea problems. We believe that this is the first time that a purely image-based learning has been trained to solve geometric construction problems of this difficulty.
△ Less
Submitted 27 June, 2021;
originally announced June 2021.
-
Vertex and edge metric dimensions of unicyclic graphs
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
In a graph G, the cardinality of the smallest ordered set of vertices that distinguishes every element of V (G) (resp. E(G)) is called the vertex (resp. edge) metric dimension of G. In [16] it was shown that both vertex and edge metric dimension of a unicyclic graph G always take values from just two explicitly given consecutive integers that are derived from the structure of the graph. A natural…
▽ More
In a graph G, the cardinality of the smallest ordered set of vertices that distinguishes every element of V (G) (resp. E(G)) is called the vertex (resp. edge) metric dimension of G. In [16] it was shown that both vertex and edge metric dimension of a unicyclic graph G always take values from just two explicitly given consecutive integers that are derived from the structure of the graph. A natural problem that arises is to determine under what conditions these dimensions take each of the two possible values. In this paper for each of these two metric dimensions we characterize three graph configurations and prove that it takes the greater of the two possible values if and only if the graph contains at least one of these configurations. One of these configurations is the same for both dimensions, while the other two are specific for each of them. This enables us to establish the exact value of the metric dimensions for a unicyclic graph and also to characterize when each of these two dimensions is greater than the other one.
△ Less
Submitted 1 April, 2021;
originally announced April 2021.
-
Extremal mixed metric dimension with respect to the cyclomatic number
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
In a graph G, the cardinality of the smallest ordered set of vertices that distinguishes every element of V(G)U E(G) is called the mixed metric dimension of G, and it is denoted by mdim(G). In [12] it was conjectured that for a graph G with cyclomatic number c(G) it holds that mdim(G) <= L1(G) + 2c(G) where L1(G) is the number of leaves in G. It is already proven that the equality holds for all tr…
▽ More
In a graph G, the cardinality of the smallest ordered set of vertices that distinguishes every element of V(G)U E(G) is called the mixed metric dimension of G, and it is denoted by mdim(G). In [12] it was conjectured that for a graph G with cyclomatic number c(G) it holds that mdim(G) <= L1(G) + 2c(G) where L1(G) is the number of leaves in G. It is already proven that the equality holds for all trees and more generally for graphs with edge-disjoint cycles in which every cycle has precisely one vertex of degree >= 3. In this paper we determine that for every Theta graph G, the mixed metric dimension mdim(G) equals 3 or 4, with 4 being attained if and only if G is a balanced Theta graph. Thus, for balanced Theta graphs the above inequality is also tight. We conclude the paper by further conjecturing that there are no other graphs, besides the ones mentioned here, for which the equality mdim(G) = L1(G) + 2c(G) holds.
△ Less
Submitted 15 December, 2020;
originally announced December 2020.
-
Mixed metric dimension of graphs with edge disjoint cycles
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
In a graph G, the cardinality of the smallest ordered set of vertices that distinguishes every element of V (G)[E(G) is called the mixed metric dimension of G. In this paper we first establish the exact value of the mixed metric dimension of a unicycic graph G which is derived from the structure of G. We further consider graphs G with edge disjoint cycles in which a unicyclic restriction Gi is int…
▽ More
In a graph G, the cardinality of the smallest ordered set of vertices that distinguishes every element of V (G)[E(G) is called the mixed metric dimension of G. In this paper we first establish the exact value of the mixed metric dimension of a unicycic graph G which is derived from the structure of G. We further consider graphs G with edge disjoint cycles in which a unicyclic restriction Gi is introduced for each cycle Ci: Applying the result for unicyclic graph to each Gi then yields the exact value of the mixed metric dimension of such a graph G. The obtained formulas for the exact value of the mixed metric dimension yield a simple sharp upper bound on the mixed metric dimension, and we conclude the paper conjecturing that the analogous bound holds for general graphs with prescribed cyclomatic number.
△ Less
Submitted 20 October, 2020;
originally announced October 2020.
-
Bounds on metric dimensions of graphs with edge disjoint cycles
Authors:
Jelena Sedlar,
Riste Škrekovski
Abstract:
In a graph G, cardinality of the smallest ordered set of vertices that distinguishes every element of V (G) is the (vertex) metric dimension of G. Similarly, the cardinality of such a set is the edge metric dimension of G, if it distinguishes E(G). In this paper these invariants are considered first for unicyclic graphs, and it is shown that the vertex and edge metric dimensions obtain values from…
▽ More
In a graph G, cardinality of the smallest ordered set of vertices that distinguishes every element of V (G) is the (vertex) metric dimension of G. Similarly, the cardinality of such a set is the edge metric dimension of G, if it distinguishes E(G). In this paper these invariants are considered first for unicyclic graphs, and it is shown that the vertex and edge metric dimensions obtain values from two particular consecutive integers, which can be determined from the structure of the graph. In particular, as a consequence, we obtain that these two invariants can differ for at most one for a same unicyclic graph. Next we extend the results to graphs with edge disjoint cycles showing that the two invariants can differ at most by c, where c is the number of cycles in such a graph. We conclude the paper with a conjecture that generalizes the previously mentioned consequences to graphs with prescribed cyclomatic number c by claiming that the difference of the invariant is still bounded by c.
△ Less
Submitted 20 October, 2020;
originally announced October 2020.
-
Is This The Right Place? Geometric-Semantic Pose Verification for Indoor Visual Localization
Authors:
Hajime Taira,
Ignacio Rocco,
Jiri Sedlar,
Masatoshi Okutomi,
Josef Sivic,
Tomas Pajdla,
Torsten Sattler,
Akihiko Torii
Abstract:
Visual localization in large and complex indoor scenes, dominated by weakly textured rooms and repeating geometric patterns, is a challenging problem with high practical relevance for applications such as Augmented Reality and robotics. To handle the ambiguities arising in this scenario, a common strategy is, first, to generate multiple estimates for the camera pose from which a given query image…
▽ More
Visual localization in large and complex indoor scenes, dominated by weakly textured rooms and repeating geometric patterns, is a challenging problem with high practical relevance for applications such as Augmented Reality and robotics. To handle the ambiguities arising in this scenario, a common strategy is, first, to generate multiple estimates for the camera pose from which a given query image was taken. The pose with the largest geometric consistency with the query image, e.g., in the form of an inlier count, is then selected in a second stage. While a significant amount of research has concentrated on the first stage, there is considerably less work on the second stage. In this paper, we thus focus on pose verification. We show that combining different modalities, namely appearance, geometry, and semantics, considerably boosts pose verification and consequently pose accuracy. We develop multiple hand-crafted as well as a trainable approach to join into the geometric-semantic verification and show significant improvements over state-of-the-art on a very challenging indoor dataset.
△ Less
Submitted 2 September, 2019; v1 submitted 13 August, 2019;
originally announced August 2019.
-
Estimating 3D Motion and Forces of Person-Object Interactions from Monocular Video
Authors:
Zongmian Li,
Jiri Sedlar,
Justin Carpentier,
Ivan Laptev,
Nicolas Mansard,
Josef Sivic
Abstract:
In this paper, we introduce a method to automatically reconstruct the 3D motion of a person interacting with an object from a single RGB video. Our method estimates the 3D poses of the person and the object, contact positions, and forces and torques actuated by the human limbs. The main contributions of this work are three-fold. First, we introduce an approach to jointly estimate the motion and th…
▽ More
In this paper, we introduce a method to automatically reconstruct the 3D motion of a person interacting with an object from a single RGB video. Our method estimates the 3D poses of the person and the object, contact positions, and forces and torques actuated by the human limbs. The main contributions of this work are three-fold. First, we introduce an approach to jointly estimate the motion and the actuation forces of the person on the manipulated object by modeling contacts and the dynamics of their interactions. This is cast as a large-scale trajectory optimization problem. Second, we develop a method to automatically recognize from the input video the position and timing of contacts between the person and the object or the ground, thereby significantly simplifying the complexity of the optimization. Third, we validate our approach on a recent MoCap dataset with ground truth contact forces and demonstrate its performance on a new dataset of Internet videos showing people manipulating a variety of tools in unconstrained environments.
△ Less
Submitted 17 June, 2019; v1 submitted 4 April, 2019;
originally announced April 2019.
-
Teaching robots to imitate a human with no on-teacher sensors. What are the key challenges?
Authors:
Radoslav Skoviera,
Karla Stepanova,
Michael Tesar,
Gabriela Sejnova,
Jiri Sedlar,
Michal Vavrecka,
Robert Babuska,
Josef Sivic
Abstract:
In this paper, we consider the problem of learning object manipulation tasks from human demonstration using RGB or RGB-D cameras. We highlight the key challenges in capturing sufficiently good data with no tracking devices - starting from sensor selection and accurate 6DoF pose estimation to natural language processing. In particular, we focus on two showcases: gluing task with a glue gun and simp…
▽ More
In this paper, we consider the problem of learning object manipulation tasks from human demonstration using RGB or RGB-D cameras. We highlight the key challenges in capturing sufficiently good data with no tracking devices - starting from sensor selection and accurate 6DoF pose estimation to natural language processing. In particular, we focus on two showcases: gluing task with a glue gun and simple block-stacking with variable blocks. Furthermore, we discuss how a linguistic description of the task could help to improve the accuracy of task description. We also present the whole architecture of our transfer of the imitated task to the simulated and real robot environment.
△ Less
Submitted 24 January, 2019;
originally announced January 2019.
-
On inverse Wiener interval problem of trees
Authors:
Jelena Sedlar
Abstract:
The Wiener index W(G) of a simple connected graph G is defined as the sum of distances over all pairs of vertices in a graph. We denote by W[T_{n}] the set of all values of Wiener index for a graph from class T_{n} of trees on n vertices. The largest interval of contiguous integers (contiguous even integers in case of odd n) is denoted by W^{int}[T_{n}]. In this paper we prove that both sets are o…
▽ More
The Wiener index W(G) of a simple connected graph G is defined as the sum of distances over all pairs of vertices in a graph. We denote by W[T_{n}] the set of all values of Wiener index for a graph from class T_{n} of trees on n vertices. The largest interval of contiguous integers (contiguous even integers in case of odd n) is denoted by W^{int}[T_{n}]. In this paper we prove that both sets are of the cardinality (1/6)n^3+O(n^2) in the case of even n, while in the case of odd n we prove that the cardinality of both sets equals (1/(12))n^3+O(n^2) solving thus two conjectures posed in literature.
△ Less
Submitted 4 April, 2017;
originally announced April 2017.
-
Predicting 1p19q Chromosomal Deletion of Low-Grade Gliomas from MR Images using Deep Learning
Authors:
Zeynettin Akkus,
Issa Ali,
Jiri Sedlar,
Timothy L. Kline,
Jay P. Agrawal,
Ian F. Parney,
Caterina Giannini,
Bradley J. Erickson
Abstract:
Objective: Several studies have associated codeletion of chromosome arms 1p/19q in low-grade gliomas (LGG) with positive response to treatment and longer progression free survival. Therefore, predicting 1p/19q status is crucial for effective treatment planning of LGG. In this study, we predict the 1p/19q status from MR images using convolutional neural networks (CNN), which could be a noninvasive…
▽ More
Objective: Several studies have associated codeletion of chromosome arms 1p/19q in low-grade gliomas (LGG) with positive response to treatment and longer progression free survival. Therefore, predicting 1p/19q status is crucial for effective treatment planning of LGG. In this study, we predict the 1p/19q status from MR images using convolutional neural networks (CNN), which could be a noninvasive alternative to surgical biopsy and histopathological analysis. Method: Our method consists of three main steps: image registration, tumor segmentation, and classification of 1p/19q status using CNN. We included a total of 159 LGG with 3 image slices each who had biopsy-proven 1p/19q status (57 nondeleted and 102 codeleted) and preoperative postcontrast-T1 (T1C) and T2 images. We divided our data into training, validation, and test sets. The training data was balanced for equal class probability and then augmented with iterations of random translational shift, rotation, and horizontal and vertical flips to increase the size of the training set. We shuffled and augmented the training data to counter overfitting in each epoch. Finally, we evaluated several configurations of a multi-scale CNN architecture until training and validation accuracies became consistent. Results: The results of the best performing configuration on the unseen test set were 93.3% (sensitivity), 82.22% (specificity), and 87.7% (accuracy). Conclusion: Multi-scale CNN with their self-learning capability provides promising results for predicting 1p/19q status noninvasively based on T1C and T2 images. Significance: Predicting 1p/19q status noninvasively from MR images would allow selecting effective treatment strategies for LGG patients without the need for surgical biopsy.
△ Less
Submitted 21 November, 2016;
originally announced November 2016.
-
Extremal unicyclic graphs with respect to additively weighted Harary index
Authors:
Jelena Sedlar
Abstract:
In this paper we define cycle-star graph CSk,n-k to be a graph on n vertices consisting of the cycle of length k and n-k leafs appended to the same vertex of the cycle. Also, we define cycle-path graph CPk,n-k to be a graph on n vertices consisting of the cycle of length k and of path on n-k vertices whose one end is linked to a vertex on a cycle. We establish that cycle-star graph CS3,n-3 is the…
▽ More
In this paper we define cycle-star graph CSk,n-k to be a graph on n vertices consisting of the cycle of length k and n-k leafs appended to the same vertex of the cycle. Also, we define cycle-path graph CPk,n-k to be a graph on n vertices consisting of the cycle of length k and of path on n-k vertices whose one end is linked to a vertex on a cycle. We establish that cycle-star graph CS3,n-3 is the only maximal graph with respect to additively weighted Harary index among all unicyclic graphs on n vertices, while cycle-path graph CP3,n-3 is the only minimal unicyclic graph (here n must be at least 5). The values of additively weighted Harary index for extremal unicyclic graphs are established, so these values are the upper and the lower bound for the value of additively weighted Harary index on the class of unicyclic graphs on n vertices.
△ Less
Submitted 18 June, 2013;
originally announced June 2013.
-
Remoteness, proximity and few other distance invariants in graphs
Authors:
Jelena Sedlar
Abstract:
We establish maximal trees and graphs for the difference of average distance and proximity proving thus the corresponding conjecture posed in [4]. We also establish maximal trees for the difference of average eccentricity and remoteness and minimal trees for the difference of remoteness and radius proving thus that the corresponding conjectures posed in [4] hold for trees.
We establish maximal trees and graphs for the difference of average distance and proximity proving thus the corresponding conjecture posed in [4]. We also establish maximal trees for the difference of average eccentricity and remoteness and minimal trees for the difference of remoteness and radius proving thus that the corresponding conjectures posed in [4] hold for trees.
△ Less
Submitted 5 April, 2012;
originally announced April 2012.
-
On augmented eccentric connectivity index of graphs and trees
Authors:
Jelena Sedlar
Abstract:
In this paper we establish all extremal graphs with respect to augmented eccentric connectivity index among all (simple connected) graphs, among trees and among trees with perfect matching. For graphs that turn out to be extremal explicit formulas for the value of augmented eccentric connectivity index are derived.
In this paper we establish all extremal graphs with respect to augmented eccentric connectivity index among all (simple connected) graphs, among trees and among trees with perfect matching. For graphs that turn out to be extremal explicit formulas for the value of augmented eccentric connectivity index are derived.
△ Less
Submitted 12 July, 2011;
originally announced July 2011.
-
Independent sets in chain cacti
Authors:
Jelena Sedlar
Abstract:
In this paper chain cacti are considered. First, for two specific classes of chain cacti (orto-chains and meta-chains of cycles with h vertices) the recurrence relation for independence polynomial is derived. That recurrence relation is then used in deriving explicit expressions for independence number and number of maximum independent sets for such chains. Also, the recurrence relation for total…
▽ More
In this paper chain cacti are considered. First, for two specific classes of chain cacti (orto-chains and meta-chains of cycles with h vertices) the recurrence relation for independence polynomial is derived. That recurrence relation is then used in deriving explicit expressions for independence number and number of maximum independent sets for such chains. Also, the recurrence relation for total number of independent sets for such graphs is derived. Finaly, the proof is provided that orto-chains and meta-chains are the only extremal chain cacti with respect to total number of independent sets (orto-chains minimal and meta-chains maximal).
△ Less
Submitted 10 May, 2011;
originally announced May 2011.
