arXiv subjects

Ken-ichi Kawarabayashi

Publications and source records attributed to Ken-ichi Kawarabayashi.

At least 73 records · Page 4Linked to original sources

arXiv2018

Adapting Local Sequential Algorithms to the Distributed Setting

It is a well known fact that sequential algorithms which exhibit a strong "local" nature can be adapted to the distributed setting given a legal graph coloring. The running time of the distributed algorithm will then be at least the number of colors. Surprisingly, this well known idea was never formally stated as a unified framework. In this paper we aim to define a robust family of local sequential algorithms which can be easily adapted to the distributed setting. We then develop new tools to further enhance these algorithms, achieving state of the art results for fundamental problems. We define a simple class of greedy-like algorithms which we call \emph{orderless-local} algorithms. We show that given a legal $c$-coloring of the graph, every algorithm in this family can be converted into a distributed algorithm running in $O(c)$ communication rounds in the CONGEST model. We show that this family is indeed robust as both the method of conditional expectations and the unconstrained submodular maximization algorithm of Buchbinder \etal \cite{BuchbinderFNS15} can be expressed as orderless-local algorithms for \emph{local utility functions} --- Utility functions which have a strong local nature to them. We use the above algorithms as a base for new distributed approximation algorithms for the weighted variants of some fundamental problems: Max $k$-Cut, Max-DiCut, Max 2-SAT and correlation clustering. We develop algorithms which have the same approximation guarantees as their sequential counterparts, up to a constant additive $ε$ factor, while achieving an $O(\log^* n)$ running time for deterministic algorithms and $O(ε^{-1})$ running time for randomized ones. This improves exponentially upon the currently best known algorithms.

cs.DC↗

arXiv2018

ClassiNet -- Predicting Missing Features for Short-Text Classification

The fundamental problem in short-text classification is \emph{feature sparseness} -- the lack of feature overlap between a trained model and a test instance to be classified. We propose \emph{ClassiNet} -- a network of classifiers trained for predicting missing features in a given instance, to overcome the feature sparseness problem. Using a set of unlabeled training instances, we first learn binary classifiers as feature predictors for predicting whether a particular feature occurs in a given instance. Next, each feature predictor is represented as a vertex $v_i$ in the ClassiNet where a one-to-one correspondence exists between feature predictors and vertices. The weight of the directed edge $e_{ij}$ connecting a vertex $v_i$ to a vertex $v_j$ represents the conditional probability that given $v_i$ exists in an instance, $v_j$ also exists in the same instance. We show that ClassiNets generalize word co-occurrence graphs by considering implicit co-occurrences between features. We extract numerous features from the trained ClassiNet to overcome feature sparseness. In particular, for a given instance $\vec{x}$, we find similar features from ClassiNet that did not appear in $\vec{x}$, and append those features in the representation of $\vec{x}$. Moreover, we propose a method based on graph propagation to find features that are indirectly related to a given short-text. We evaluate ClassiNets on several benchmark datasets for short-text classification. Our experimental results show that by using ClassiNet, we can statistically significantly improve the accuracy in short-text classification tasks, without having to use any external resources such as thesauri for finding related features.

cs.CL↗

arXiv2017

Think Globally, Embed Locally --- Locally Linear Meta-embedding of Words

Distributed word embeddings have shown superior performances in numerous Natural Language Processing (NLP) tasks. However, their performances vary significantly across different tasks, implying that the word embeddings learnt by those methods capture complementary aspects of lexical semantics. Therefore, we believe that it is important to combine the existing word embeddings to produce more accurate and complete \emph{meta-embeddings} of words. For this purpose, we propose an unsupervised locally linear meta-embedding learning method that takes pre-trained word embeddings as the input, and produces more accurate meta embeddings. Unlike previously proposed meta-embedding learning methods that learn a global projection over all words in a vocabulary, our proposed method is sensitive to the differences in local neighbourhoods of the individual source word embeddings. Moreover, we show that vector concatenation, a previously proposed highly competitive baseline approach for integrating word embeddings, can be derived as a special case of the proposed method. Experimental results on semantic similarity, word analogy, relation classification, and short-text classification tasks show that our meta-embeddings to significantly outperform prior methods in several benchmark datasets, establishing a new state of the art for meta-embeddings.

cs.CL↗

arXiv2017

Using $k$-way Co-occurrences for Learning Word Embeddings

Co-occurrences between two words provide useful insights into the semantics of those words. Consequently, numerous prior work on word embedding learning have used co-occurrences between two words as the training signal for learning word embeddings. However, in natural language texts it is common for multiple words to be related and co-occurring in the same context. We extend the notion of co-occurrences to cover $k(\geq\!\!2)$-way co-occurrences among a set of $k$-words. Specifically, we prove a theoretical relationship between the joint probability of $k(\geq\!\!2)$ words, and the sum of $\ell_2$ norms of their embeddings. Next, we propose a learning objective motivated by our theoretical result that utilises $k$-way co-occurrences for learning word embeddings. Our experimental results show that the derived theoretical relationship does indeed hold empirically, and despite data sparsity, for some smaller $k$ values, $k$-way embeddings perform comparably or better than $2$-way embeddings in a range of tasks.

cs.CL↗

arXiv2017

Polylogarithmic approximation for minimum planarization (almost)

In the minimum planarization problem, given some $n$-vertex graph, the goal is to find a set of vertices of minimum cardinality whose removal leaves a planar graph. This is a fundamental problem in topological graph theory. We present a $\log^{O(1)} n$-approximation algorithm for this problem on general graphs with running time $n^{O(\log n/\log\log n)}$. We also obtain a $O(n^\varepsilon)$-approximation with running time $n^{O(1/\varepsilon)}$ for any arbitrarily small constant $\varepsilon > 0$. Prior to our work, no non-trivial algorithm was known for this problem on general graphs, and the best known result even on graphs of bounded degree was a $n^{Ω(1)}$-approximation [Chekuri and Sidiropoulos 2013]. As an immediate corollary, we also obtain improved approximation algorithms for the crossing number problem on graphs of bounded degree. Specifically, we obtain $O(n^{1/2+\varepsilon})$-approximation and $n^{1/2} \log^{O(1)} n$-approximation algorithms in time $n^{O(1/\varepsilon)}$ and $n^{O(\log n/\log\log n)}$ respectively. The previously best-known result was a polynomial-time $n^{9/10}\log^{O(1)} n$-approximation algorithm [Chuzhoy 2011]. Our algorithm introduces several new tools including an efficient grid-minor construction for apex graphs, and a new method for computing irrelevant vertices. Analogues of these tools were previously available only for exact algorithms. Our work gives efficient implementations of these ideas in the setting of approximation algorithms, which could be of independent interest.

cs.DS↗

arXiv2017

Additive non-approximability of chromatic number in proper minor-closed classes

Robin Thomas asked whether for every proper minor-closed class C, there exists a polynomial-time algorithm approximating the chromatic number of graphs from C up to a constant additive error independent on the class C. We show this is not the case: unless P=NP, for every integer k>=1, there is no polynomial-time algorithm to color a K_{4k+1}-minor-free graph G using at most chi(G)+k-1 colors. More generally, for every k>=1 and 1<=β<=4/3, there is no polynomial-time algorithm to color a K_{4k+1}-minor-free graph G using less than beta.chi(G)+(4-3beta)k colors. As far as we know, this is the first non-trivial non-approximability result regarding the chromatic number in proper minor-closed classes. We also give somewhat weaker non-approximability bound for K_{4k+1}-minor-free graphs with no cliques of size 4. On the positive side, we present additive approximation algorithm whose error depends on the apex number of the forbidden minor, and an algorithm with additive error 6 under the additional assumption that the graph has no 4-cycles.

cs.DM↗

arXiv2017

Triangle-free graphs of tree-width t are ceil((t + 3)/2)-colorable

We prove that every triangle-free graph of tree-width t has chromatic number at most ceil((t + 3)/2), and demonstrate that this bound is tight. The argument also establishes a connection between coloring graphs of tree-width t and on-line coloring of graphs of path-width t.

math.CO↗

arXiv2016

Optimal Pricing for Submodular Valuations with Bounded Curvature

The optimal pricing problem is a fundamental problem that arises in combinatorial auctions. Suppose that there is one seller who has indivisible items and multiple buyers who want to purchase a combination of the items. The seller wants to sell his items for the highest possible prices, and each buyer wants to maximize his utility (i.e., valuation minus payment) as long as his payment does not exceed his budget. The optimal pricing problem seeks a price of each item and an assignment of items to buyers such that every buyer achieves the maximum utility under the prices. The goal of the problem is to maximize the total payment from buyers. In this paper, we consider the case that the valuations are submodular. We show that the problem is computationally hard even if there exists only one buyer. Then we propose approximation algorithms for the unlimited budget case. We also extend the algorithm for the limited budget case when there exists one buyer and multiple buyers collaborate with each other.

cs.GT↗

arXiv2016

A New Proof of the Flat Wall Theorem

We give an elementary and self-contained proof, and a numerical improvement, of a weaker form of the excluded clique minor theorem of Robertson and Seymour, the following. Let t,r>0 be integers, and let R=49152t^{24}(40t^2+r). An r-wall is obtained from a (2r x r)-grid by deleting every odd vertical edge in every odd row and every even vertical edge in every even row, then deleting the two resulting vertices of degree one, and finally subdividing edges arbitrarily. The vertices of degree two that existed before the subdivision are called the pegs of the r-wall. Let G be a graph with no K_t minor, and let W be an R-wall in G. We prove that there exist a subset A of V(G) of size at most 12288t^{24} and an r-subwall W' of W such that V(W') is disjoint from A and W' is a flat wall in G-A in the following sense. There exists a separation (X,Y) of G-A such that X\cap Y is a subset of the vertex set of the cycle C' that bounds the outer face of W', V(W') is a subset of Y, every peg of W' belongs to X and the graph G[Y] can almost be drawn in the unit disk with the vertices X\cap Y drawn on the boundary of the disk in the order determined by C'. Here almost means that the assertion holds after repeatedly removing parts of the graph separated from X\cap Y by a cutset Z of size at most three, and adding all edges with both ends in Z. Our proof gives rise to an algorithm that runs in polynomial time even when r and t are part of the input instance. The proof is self-contained in the sense that it uses only results whose proofs can be found in textbooks.

math.CO↗

arXiv2016

Successor-Invariant First-Order Logic on Graphs with Excluded Topological Subgraphs

We show that the model-checking problem for successor-invariant first-order logic is fixed-parameter tractable on graphs with excluded topological subgraphs when parameterised by both the size of the input formula and the size of the exluded topological subgraph. Furthermore, we show that model-checking for order-invariant first-order logic is tractable on coloured posets of bounded width, parameterised by both the size of the input formula and the width of the poset. Our result for successor-invariant FO extends previous results for this logic on planar graphs (Engelmann et al., LICS 2012) and graphs with excluded minors (Eickmeyer et al., LICS 2013), further narrowing the gap between what is known for FO and what is known for successor-invariant FO. The proof uses Grohe and Marx's structure theorem for graphs with excluded topological subgraphs. For order-invariant FO we show that Gajarský et al.'s recent result for FO carries over to order-invariant FO.

cs.LO↗

arXiv2016

A coherent Ising machine for MAX-CUT problems : Performance evaluation against semidefinite programming relaxation and simulated annealing

Combinatorial optimization problems are computationally hard in general, but they are ubiquitous in our modern life. A coherent Ising machine (CIM) based on a multiple-pulse degenerate optical parametric oscillator (DOPO) is an alternative approach to solve these problems by a specialized physical computing system. To evaluate its potential performance, computational experiments are performed on maximum cut (MAX-CUT) problems against traditional algorithms such as semidefinite programming relaxation of Goemans-Williamson and simulated annealing by Kirkpatrick, et al. The numerical results empirically suggest that the almost constant computation time is required to obtain the reasonably accurate solutions of MAX-CUT problems on a CIM with the number of vertices up to $2 \times 10^4$ and the number of edges up to $10^8$.

quant-ph↗

arXiv2015

Joint Word Representation Learning using a Corpus and a Semantic Lexicon

Methods for learning word representations using large text corpora have received much attention lately due to their impressive performance in numerous natural language processing (NLP) tasks such as, semantic similarity measurement, and word analogy detection. Despite their success, these data-driven word representation learning methods do not consider the rich semantic relational structure between words in a co-occurring context. On the other hand, already much manual effort has gone into the construction of semantic lexicons such as the WordNet that represent the meanings of words by defining the various relationships that exist among the words in a language. We consider the question, can we improve the word representations learnt using a corpora by integrating the knowledge from semantic lexicons?. For this purpose, we propose a joint word representation learning method that simultaneously predicts the co-occurrences of two words in a sentence subject to the relational constrains given by the semantic lexicon. We use relations that exist between words in the lexicon to regularize the word representations learnt from the corpus. Our proposed method statistically significantly outperforms previously proposed methods for incorporating semantic lexicons into word representations on several benchmark datasets for semantic similarity and word analogy.

cs.CL↗

arXiv2015

Graph Isomorphism for Bounded Genus Graphs In Linear Time

For every integer $g$, isomorphism of graphs of Euler genus at most $g$ can be decided in linear time. This improves previously known algorithms whose time complexity is $n^{O(g)}$ (shown in early 1980's), and in fact, this is the first fixed-parameter tractable algorithm for the graph isomorphism problem for bounded genus graphs in terms of the Euler genus $g$. Our result also generalizes the seminal result of Hopcroft and Wong in 1974, which says that the graph isomorphism problem can be decided in linear time for planar graphs. Our proof is quite lengthly and complicated, but if we are satisfied with an $O(n^3)$ time algorithm for the same problem, the proof is shorter and easier.

cs.DS↗

arXiv2015

Connectivity Preserving Iterative Compaction and Finding 2 Disjoint Rooted Paths in Linear Time

In this paper we show how to combine two algorithmic techniques to obtain linear time algorithms for various optimization problems on graphs, and present a subroutine which will be useful in doing so. The first technique is iterative shrinking. In the first phase of an iterative shrinking algorithm, we construct a sequence of graphs of decreasing size $G_1,\ldots,G_\ell$ where $G_1$ is the initial input, $G_\ell$ is a graph on which the problem is easy, and $G_i$ is obtained from $G_{i+1}$ via some shrinking algorithm. In the second phase we work through the sequence in reverse, repeatedly constructing a solution for a graph from the solution for its successor. In an iterative compaction algorithm, we insist that the graphs decrease by a constant fraction of the entire graph. Another approach to solving optimization problems is to exploit the structural properties implied by the connectivity of the input graph. This approach can be used on graphs which are not highly connected by decomposing an input graph into its highly connected pieces, solving subproblems on these specially structured pieces and then combining their solutions. We combine these two techniques by developing compaction algorithms which when applied to the highly connected pieces preserve their connectivity properties. The structural properties this connectivity implies can be helpful both in finding further compactions in later iterations and when we are manipulating solutions in the second phase of an iterative compaction algorithm. To illustrate how this compaction algorithm can be used as a subroutine, we present a linear time algorithm that given four vertices $\{s_1,s_2,t_1,t_2\}$ of a graph $G$, either finds a pair of disjoint paths $P_1$ and $P_2$ of $G$ such that $P_i$ has endpoints $s_i$ and $t_i$, or returns a planar embedding of an auxiliary graph which shows that no such pair exists.

cs.DS↗

arXiv2015

The odd Hadwiger's conjecture is "almost'' decidable

The odd Hadwiger's conjecture, made by Gerads and Seymour in early 1990s, is an analogue of the famous Hadwiger's conjecture. It says that every graph with no odd $K_t$-minor is $(t-1)$-colorable. This conjecture is known to be true for $t \leq 5$, but the cases $t \geq 5$ are wide open. So far, the most general result says that every graph with no odd $K_t$-minor is $O(t \sqrt{\log t})$-colorable. In this paper, we tackle this conjecture from an algorithmic view, and show the following: For a given graph $G$ and any fixed $t$, there is a polynomial time algorithm to output one of the following: \begin{enumerate} \item a $(t-1)$-coloring of $G$, or \item an odd $K_{t}$-minor of $G$, or \item after making all "reductions" to $G$, the resulting graph $H$ (which is an odd minor of $G$ and which has no reductions) has a tree-decomposition $(T, Y)$ such that torso of each bag $Y_t$ is either \begin{itemize} \item of size at most $f_1(t) \log n$ for some function $f_1$ of $t$, or \item a graph that has a vertex $X$ of order at most $f_2(t)$ for some function $f_2$ of $t$ such that $Y_t-X$ is bipartite. Moreover, degree of $t$ in $T$ is at most $f_3(t)$ for some function $f_3$ of $t$. \end{itemize} \end{enumerate} Let us observe that the last odd minor $H$ is indeed a minimal counterexample to the odd Hadwiger's conjecture for the case $t$. So our result says that a minimal counterexample satisfies the lsat conclusion.

math.CO↗

arXiv2015

Unsupervised Cross-Domain Word Representation Learning

Meaning of a word varies from one domain to another. Despite this important domain dependence in word semantics, existing word representation learning methods are bound to a single domain. Given a pair of \emph{source}-\emph{target} domains, we propose an unsupervised method for learning domain-specific word representations that accurately capture the domain-specific aspects of word semantics. First, we select a subset of frequent words that occur in both domains as \emph{pivots}. Next, we optimize an objective function that enforces two constraints: (a) for both source and target domain documents, pivots that appear in a document must accurately predict the co-occurring non-pivots, and (b) word representations learnt for pivots must be similar in the two domains. Moreover, we propose a method to perform domain adaptation using the learnt word representations. Our proposed method significantly outperforms competitive baselines including the state-of-the-art domain-insensitive word representations, and reports best sentiment classification accuracies for all domain-pairs in a benchmark dataset.

cs.CL↗

arXiv2015

Embedding Semantic Relations into Word Representations

Learning representations for semantic relations is important for various tasks such as analogy detection, relational search, and relation classification. Although there have been several proposals for learning representations for individual words, learning word representations that explicitly capture the semantic relations between words remains under developed. We propose an unsupervised method for learning vector representations for words such that the learnt representations are sensitive to the semantic relations that exist between two words. First, we extract lexical patterns from the co-occurrence contexts of two words in a corpus to represent the semantic relations that exist between those two words. Second, we represent a lexical pattern as the weighted sum of the representations of the words that co-occur with that lexical pattern. Third, we train a binary classifier to detect relationally similar vs. non-similar lexical pattern pairs. The proposed method is unsupervised in the sense that the lexical pattern pairs we use as train data are automatically sampled from a corpus, without requiring any manual intervention. Our proposed method statistically significantly outperforms the current state-of-the-art word representations on three benchmark datasets for proportional analogy detection, demonstrating its ability to accurately capture the semantic relations among words.

cs.CL↗

arXiv2014

Learning Word Representations from Relational Graphs

Attributes of words and relations between two words are central to numerous tasks in Artificial Intelligence such as knowledge representation, similarity measurement, and analogy detection. Often when two words share one or more attributes in common, they are connected by some semantic relations. On the other hand, if there are numerous semantic relations between two words, we can expect some of the attributes of one of the words to be inherited by the other. Motivated by this close connection between attributes and relations, given a relational graph in which words are inter- connected via numerous semantic relations, we propose a method to learn a latent representation for the individual words. The proposed method considers not only the co-occurrences of words as done by existing approaches for word representation learning, but also the semantic relations in which two words co-occur. To evaluate the accuracy of the word representations learnt using the proposed method, we use the learnt word representations to solve semantic word analogy problems. Our experimental results show that it is possible to learn better word representations by using semantic semantics between words.

cs.CL↗