Published papers

In this paper, we consider the problem of maximizing a monotone submodular function subject to a knapsack constraint in the streaming setting. In particular, the elements arrive sequentially and at any point of time, the algorithm has access only to a small fraction of the data stored in primary memory. For this problem, we propose a (0.363-ϵ)-approximation algorithm, requiring only a single pass through the data; moreover, we propose a (0.4 - ϵ)-approximation algorithm requiring a constant number of passes through the data. The required memory space of both algorithms depends only on the size of the knapsack capacity and ϵ.

© 2018 Curran Associates Inc.All rights reserved. Online portfolio selection is a sequential decision-making problem in which a learner repetitively selects a portfolio over a set of assets, aiming to maximize long-term return. In this paper, we study the problem with the cardinality constraint that the number of assets in a portfolio is restricted to be at most k, and consider two scenarios: (i) in the full-feedback setting, the learner can observe price relatives (rates of return to cost) for all assets, and (ii) in the bandit-feedback setting, the learner can observe price relatives only for invested assets. We propose efficient algorithms for these scenarios, which achieve sublinear regrets. We also provide regret (statistical) lower bounds for both scenarios which nearly match the upper bounds when k is a constant. In addition, we give a computational lower bound, which implies that no algorithm maintains both computational efficiency, as well as a small regret upper bound.

© N. Kakimura, N. Kamiyama, and K. Takazawa. In this paper, we introduce the concept of b-branchings in digraphs, which is a generalization of branchings serving as a counterpart of b-matchings. Here b is a positive integer vector on the vertex set of a digraph, and a b-branching is defined as a common independent set of two matroids defined by b: an arc set is a b-branching if it has at most b(v) arcs sharing the terminal vertex v, and it is an independent set of a certain sparsity matroid defined by b. We demonstrate that b-branchings yield an appropriate generalization of branchings by extending several classical results on branchings. We first present a multi-phase greedy algorithm for finding a maximum-weight b-branching. We then prove a packing theorem extending Edmonds’ disjoint branchings theorem, and provide a strongly polynomial algorithm for finding optimal disjoint b-branchings. As a consequence of the packing theorem, we prove the integer decomposition property of the bbranching polytope. Finally, we deal with a further generalization in which a matroid constraint is imposed on the b(v) arcs sharing the terminal vertex v.

We consider set covering problems where the underlying set system satisfies a particular replacement property w.r.t. a given partial order on the elements: Whenever a set is in the set system then a set stemming from it via the replacement of an element by a smaller element is also in the set system. Many variants of Bin Packing that have appeared in the literature are such set covering problems with ordered replacement. We provide a rigorous account on the additive and multiplicative integrality gap and approximability of set covering with replacement. In particular we provide a polylogarithmic upper bound on the additive integrality gap that also yields a polynomial time additive approximation algorithm if the linear programming relaxation can be efficiently solved. We furthermore present an extensive list of covering problems that fall into our framework and consequently have polylogarithmic additive gaps as well.

We consider the problem of reforming an envy-free matching when each agent is assigned a single item. Given an envy-free matching, we consider an operation to exchange the item of an agent with an unassigned item preferred by the agent that results in another envy-free matching. We repeat this operation as long as we can. We prove that the resulting envy-free matching is uniquely determined up to the choice of an initial envy-free matching, and can be found in polynomial time. We call the resulting matching a reformist envy-free matching, and then we study a shortest sequence to obtain the reformist envy-free matching from an initial envy-free matching. We prove that a shortest sequence is computationally hard to obtain even when each agent accepts at most four items and each item is accepted by at most three agents. On the other hand, we give polynomial-time algorithms when each agent accepts at most three items or each item is accepted by at most two agents. Inapproximability and fixed-parameter (in)tractability are also discussed.

Copyright 2018 by the author(s). We investigate an online regression problem in which the learner makes predictions sequentially while only the limited information on features is observable. In this paper, we propose a general setting for the limitation of the available information, where the observed information is determined by a function chosen from a given set of observation functions. Our problem setting is a generalization of the online sparse linear regression problem, which has been actively studied. For our general problem, we present an algorithm by combining multi-armed bandit algorithms and online learning methods. This algorithm admits a sublinear regret bound when the number of observation functions is constant. We also show that the dependency on the number of observation functions is inevitable unless additional assumptions are adopted. To mitigate this inefficiency, we focus on a special case of practical importance, in which the observed information is expressed through linear combinations of the original features. We propose efficient algorithms for this special case. Finally, we also demonstrate the efficiency of the proposed algorithms by simulation studies using both artificial and real data.

In this paper, we propose new exact and approximation algorithms for the weighted matroid intersection problem. Our exact algorithm is faster than previous algorithms when the largest weight is relatively small. Our approximation algorithm delivers a (1 - ϵ)-approximate solution with a running time significantly faster than known exact algorithms. The core of our algorithms is a decomposition technique: we decompose an instance of the weighted matroid intersection problem into a set of instances of the unweighted matroid intersection problem. The computational advantage of this approach is that we can make use of fast unweighted matroid intersection algorithms as a black box for designing algorithms. Precisely speaking, we prove that we can solve the weighted matroid intersection problem via solving W instances of the unweighted matroid intersection problem, where W is the largest given weight. Furthermore, we can find a (1 - ϵ)-approximate solution via solving O(ϵ-1 log r) instances of the unweighted matroid intersection problem, where r is the smallest rank of the given two matroids. Our algorithms are simple and flexible: they can be adapted to special cases of the weighted matroid intersection problem, using specialized unweighted matroid intersection algorithms. In this paper, we will show the following results. 1. Given two general matroids, using Cunningham's algorithm, we can solve the weighted matroid intersection problem exactly in O(τWnr1.5) time and (1 - ϵ)-approximately in O(τϵ-1nr1.5 logr) time, where n is the size of the ground set and r is the time complexity of an independence oracle call. 2. Given two graphic matroids, using the algorithm of Gabow and Xu, we can solve the weighted matroid intersection problem exactly in O(W√rn log r) time and (1 - ϵ)-approximately in O(ϵ-1 √rn log2r) time. 3. Given two linear matroids (in the form of two r-by-n matrices), using the algorithm of Cheung, Kwok, and Lau, we can solve the weighted matroid intersection problem exactly in O(nr log r∗ + Wnrω∗-1) time and (1 - ϵ)-approximately in O(nr logr∗ + ϵ-1nrω∗-1 logr∗) time, where ω is the exponent of the matrix multiplication time and r∗ is the maximum size of a common independent set. Finally, we give a further application of our decomposition technique. We use our technique to solve efficiently the rankmaximal matroid intersection problem, a problem motivated by matching problems under preferences.

© Springer Nature Switzerland AG 2019. In this paper, we consider the problem of maximizing a monotone submodular function subject to a knapsack constraint in the streaming setting. In particular, the elements arrive sequentially and at any point of time, the algorithm has access only to a small fraction of the data stored in primary memory. For this problem, we propose a (0.4 - ε) -approximation algorithm requiring only a single pass through the data. This improves on the currently best (0.363 - ε) -approximation algorithm. The required memory space depends only on the size of the knapsack capacity and ε.

© 2017 Neural information processing systems foundation. All rights reserved. Online sparse linear regression is the task of applying linear regression analysis to examples arriving sequentially subject to a resource constraint that a limited number of features of examples can be observed. Despite its importance in many practical applications, it has been recently shown that there is no polynomialtime sublinear-regret algorithm unless NP ⊆ BPP, and only an exponential-time sublinear-regret algorithm has been found. In this paper, we introduce mild assumptions to solve the problem. Under these assumptions, we present polynomialtime sublinear-regret algorithms for the online sparse linear regression. In addition, thorough experiments with publicly available data demonstrate that our algorithms outperform other known algorithms.

© 35th International Conference on Machine Learning, ICML 2018.All Rights Reserved. Bandit is a framework for designing sequential experiments, where a learner selects an arm A ϵ A and obtains an observation corresponding to A in each experiment. Theoretically, the tight regret lower-bound for the general bandit is polynomial with respect to the number of arms |A|, and thus, to overcome this bound, the bandit problem with side-information is often considered. Recently, a bandit framework over a causal graph was introduced, where the structure of the causal graph is available as side-information and the arms are identified with interventions on the causal graph. Existing algorithms for causal bandit overcame the Ω(√\A\/T) simple-regret lower-bound; however, their algorithms work only when the interventions A are localized around a single node (i.e., an intervention propagates only to its neighbors). We then propose a novel causal bandit algorithm for an arbitrary set of interventions, which can propagate throughout the causal graph. We also show that it achieves O(√γ∗ log(|A|T)/T) regret bound, where γ∗ is determined by using a causal graph structure. In particular, if the maximum in-degree of the causal graph is a constant, then γ∗ = O(N2), where N is the number of nodes.

© 2017 IEEE. We aim to decrease a communication cost of a network that uses compressive sensing, a technique that allows us to recover global information of sparse data by using only a small set of samples. Despite efficiency of the technique, collecting information from all samples is usually costly. Because the samples from previous works usually spread around the network, setting up a number of base stations does not help reducing the cost. In this paper, we propose a method that can utilize the base stations, while aiming to minimize the recovery error of compressive sensing. Based on theorem by Xu et al., which is for cost-aware compressive sensing, we derive a mathematical program that aims to maximize the preciseness in the setting. Then, we approximate the program by a convex quadratic program and prove that the approximation ratio is 0.63. Our simulation results show that, by using the coverage, the sampling error is decreased by at most thirty times.

This paper focuses on a generalization of the traveling salesman problem (TSP), called the subpath planning problem (SPP). Given 2n vertices and n independent edges on a metric space, we aim to find a shortest tour that contains all the edges. SPP is one of the fundamental problems in both artificial intelligence and robotics. Our main result is to design a 1.5-approximation algorithm that runs in polynomial time, improving the currently best approximation algorithm. The idea is direct use of techniques developed for TSP. In addition, we propose a generalization of SPP called the subgroup planning problem (SGPP). In this problem, we are given a set of disjoint groups of vertices, and we aim to find a shortest tour such that all the vertices in each group are traversed sequentially. We propose a 3-approximation algorithm for SGPP. We also conduct numerical experiments. Compared with previous algorithms, our algorithms improve the solution quality by more than 10% for large instances with more than 10,000 vertices.

The multi-service center problem is a variant of facility location problems. In the problem, we consider locating p facilities on a graph, each of which provides distinct service required by all vertices. Each vertex incurs the cost determined by the sum of the weighted distances to the p facilities. The aim of the problem is to minimize the maximum cost among all vertices. This problem is known to be NP-hard for general graphs, while it is solvable in polynomial time when p is a fixed constant. In this paper, we give sharp analyses for the complexity of the problem from the viewpoint of graph classes and weights on vertices. We first propose a polynomial-Time algorithm for trees when p is a part of input. In contrast, we prove that the problem becomes strongly NP-hard even for cycles. We also show that when vertices are allowed to have negative weights, the problem becomes NP-hard for paths of only three vertices and strongly NP-hard for stars.

Influence maximization is a well-studied problem of finding a small set of highly influential individuals in a social network such that the spread of influence under a certain diffusion model is maximized. We propose new diffusion models that incorporate the time-decaying phenomenon by which the power of influence decreases with elapsed time. In standard diffusion models such as the independent cascade and linear threshold models, each edge in a network has a fixed power of influence over time. However, in practical settings, such as rumor spreading, it is natural for the power of influence to depend on the time influenced. We generalize the independent cascade and linear threshold models with time-decaying effects. Moreover, we show that by using an analysis framework based on submodular functions, a natural greedy strategy obtains a solution that is provably within (1 − 1/e) of optimal. In addition, we propose theoretically and practically fast algorithms for the proposed models. Experimental results show that the proposed algorithms are scalable to graphs with millions of edges and outperform baseline algorithms based on a state-of-the-art algorithm.

© 2018, Springer Nature Switzerland AG. We consider and formulate problems of PageRank score boosting motivated by applications such as effective web advertising. More precisely, given a graph and target vertices, one is required to find a fixed-size set of missing edges that maximizes the minimum PageRank score among the targets. We provide theoretical analyses to show that all of them are NP-hard. To overcome the hardness, we develop heuristic-based algorithms for them. We finally perform experiments on several real-world networks to verify the effectiveness of the proposed algorithms compared to baselines. Specifically, our algorithm achieves 100 times improvements of the minimum PageRank score among selected 100 vertices by adding only dozens of edges.

© 2018 Association for Computing Machinery. Community detection is one of the fundamental tasks in graph mining, which has many real-world applications in diverse domains. In this study, we propose an optimization model for finding a community that is densely connected internally but sparsely connected to the rest of the graph. The model extends the densest subgraph problem, in which we maximize the density while minimizing the average cut size. We first show that our proposed model can be solved efficiently. Then we design two polynomial-time exact algorithms based on linear programming and a maximum flow algorithm, respectively. Moreover, to deal with larger-sized graphs in practice, we present a scalable greedy algorithm that runs in almost linear time with theoretical performance guarantee of the output. In addition, as our model is closely related to a quality function called the modularity density, we show that our algorithms can also be used to find global community structure in a graph. With thorough experiments using well-known real-world graphs, we demonstrate that our algorithms are highly effective in finding a suitable community in a graph. For example, for web-Google, our algorithm finds a solution with more than 99.1% density and less than 3.1% cut size, compared with a solution obtained by a baseline algorithm for the densest subgraph problem.

© 2014 IEEE. Scheduling sensors to prolong the lifetime of covering targets in the field is one of the central problems in wireless sensor networks. This problem, called the maximum lifetime coverage problem (MLCP), can be formulated as a linear programming problem with exponential size, and has a constant-factor approximation algorithm. In reality, however, batteries of sensors have recovery effects, which is a phenomenon that the deliverable energy in batteries can be replenished by itself if it is left idling for sufficient duration. Thanks to that effects, we can obtain much longer lifetime of sensors if each sensor is forced to take a sleep at some interval. In this paper, we introduce two models that extend the MLCP, incorporating battery recovery effects. The first model represents battery recovery effects in a deterministic way, while the second one uses a probabilistic model to imitate the effects. We then propose efficient algorithms that work for both models by extending approximation algorithms for the original MLCP. Numerical experiments show that the lifetime of our schedule is 10-40% longer than one without battery recovery effects.

2014 We consider the budget allocation problem over bipartite influence model proposed by Alon et al. (Alon et al., 2012). This problem can be viewed as the well-known influence maximization problem with budget constraints. We first show that this problem and its much more general form fall into a general setting; namely the monotone submodular function maximization over integer lattice subject to a knapsack constraint. Our framework includes Alon et al.'s model, even with a competitor and with cost. We then give a (1 - 1/e)-approximation algorithm for this more general problem. Furthermore, when influence probabilities are nonincreasing, we obtain a faster (1 - 1/e)-approximation algorithm, which runs essentially in linear time in the number of nodes. This allows us to implement our algorithm up to almost 10M edges (indeed, our experiments tell us that we can implement our algorithm up to 1 billion edges. It would approximately take us only 500 seconds.).

Copyright © 2015 by the author(s). We propose a new influence model for allocating budgets to advertising channels. Our model captures customer's sensitivity to advertisements as a threshold behavior; a customer is expected to be influenced if the influence he receives exceeds his threshold. Over the threshold model, we discuss two optimization problems. The first one is the budget-constrained influence maximization. We propose two greedy algorithms based on different strategies, and analyze the performance when the influence is submodular. We then introduce a new characteristic to measure the cost-effectiveness of a marketing campaign, that is, the proportion of the resulting influence to the cost spent. We design an almost linear-time approximation algorithm to maximize the cost-effectiveness. Furthermore, we design a better-approximation algorithm based on linear programming for a special case. We conduct thorough experiments to confirm that our algorithms outperform baseline algorithms.

© Springer International Publishing AG 2018. The server allocation problem is a facility location problem for a distributed processing scheme on a real-time network. In this problem, we are given a set of users and a set of servers. Then, we consider the following communication process between users and servers. First a user sends his/her request to the nearest server. After receiving all the requests from users, the servers share the requests. A user will then receive the data processed from the nearest server. The goal of this problem is to choose a subset of servers so that the total delay of the above process is minimized. In this paper, we prove the following approximability and inapproximability results. We first show that the server allocation problem has no polynomial-time approximation algorithm unless P = NP. However, assuming that the delays satisfy the triangle inequality, we design a polynomial-time 3/2-approximation algorithm. When we assume the triangle inequality only among servers, we propose a polynomial-time 2-approximation algorithm. Both of the algorithms are tight in the sense that we cannot obtain better polynomial-time approximation algorithms unless P = NP. Furthermore, we evaluate the practical performance of our algorithms through computational experiments, and show that our algorithms scale better and produce comparable solutions than the previously proposed method based on integer linear programming.