arXiv subjects

Naoto Ohsaka

Publications and source records attributed to Naoto Ohsaka.

28 records · Page 2Linked to original sources

arXiv2023

Curse of "Low" Dimensionality in Recommender Systems

Beyond accuracy, there are a variety of aspects to the quality of recommender systems, such as diversity, fairness, and robustness. We argue that many of the prevalent problems in recommender systems are partly due to low-dimensionality of user and item embeddings, particularly when dot-product models, such as matrix factorization, are used. In this study, we showcase empirical evidence suggesting the necessity of sufficient dimensionality for user/item embeddings to achieve diverse, fair, and robust recommendation. We then present theoretical analyses of the expressive power of dot-product models. Our theoretical results demonstrate that the number of possible rankings expressible under dot-product models is exponentially bounded by the dimension of item factors. We empirically found that the low-dimensionality contributes to a popularity bias, widening the gap between the rank positions of popular and long-tail items; we also give a theoretical justification for this phenomenon.

cs.IR↗

arXiv2022

On Reconfigurability of Target Sets

We study the problem of deciding reconfigurability of target sets of a graph. Given a graph $G$ with vertex thresholds $τ$, consider a dynamic process in which vertex $v$ becomes activated once at least $τ(v)$ of its neighbors are activated. A vertex set $S$ is called a target set if all vertices of $G$ would be activated when initially activating vertices of $S$. In the Target Set Reconfiguration problem, given two target sets $X$ and $Y$ of the same size, we are required to determine whether $X$ can be transformed into $Y$ by repeatedly swapping one vertex in the current set with another vertex not in the current set preserving every intermediate set as a target set. In this paper, we investigate the complexity of Target Set Reconfiguration in restricted cases. On the hardness side, we prove that Target Set Reconfiguration is PSPACE-complete on bipartite planar graphs of degree $3$ and $4$ and of threshold $2$, bipartite $3$-regular graphs and planar $3$-regular graphs of threshold $1$ and $2$, and split graphs, which is in contrast to the fact that a special case called Vertex Cover Reconfiguration is in P for the last graph class. On the positive side, we present a polynomial-time algorithm for Target Set Reconfiguration on graphs of maximum degree $2$ and trees. The latter result can be thought of as a generalization of that for Vertex Cover Reconfiguration.

cs.DS↗

arXiv2022

Some Inapproximability Results of MAP Inference and Exponentiated Determinantal Point Processes

We study the computational complexity of two hard problems on determinantal point processes (DPPs). One is maximum a posteriori (MAP) inference, i.e., to find a principal submatrix having the maximum determinant. The other is probabilistic inference on exponentiated DPPs (E-DPPs), which can sharpen or weaken the diversity preference of DPPs with an exponent parameter $p$. We present several complexity-theoretic hardness results that explain the difficulty in approximating MAP inference and the normalizing constant for E-DPPs. We first prove that unconstrained MAP inference for an $n \times n$ matrix is $\textsf{NP}$-hard to approximate within a factor of $2^{βn}$, where $β= 10^{-10^{13}} $. This result improves upon the best-known inapproximability factor of $(\frac{9}{8}-ε)$, and rules out the existence of any polynomial-factor approximation algorithm assuming $\textsf{P} \neq \textsf{NP}$. We then show that log-determinant maximization is $\textsf{NP}$-hard to approximate within a factor of $\frac{5}{4}$ for the unconstrained case and within a factor of $1+10^{-10^{13}}$ for the size-constrained monotone case. In particular, log-determinant maximization does not admit a polynomial-time approximation scheme unless $\textsf{P} = \textsf{NP}$. As a corollary of the first result, we demonstrate that the normalizing constant for E-DPPs of any (fixed) constant exponent $p \geq β^{-1} = 10^{10^{13}}$ is $\textsf{NP}$-hard to approximate within a factor of $2^{βpn}$, which is in contrast to the case of $p \leq 1$ admitting a fully polynomial-time randomized approximation scheme.

cs.DS↗

arXiv2021

Computational Complexity of Normalizing Constants for the Product of Determinantal Point Processes

We consider the product of determinantal point processes (DPPs), a point process whose probability mass is proportional to the product of principal minors of multiple matrices, as a natural, promising generalization of DPPs. We study the computational complexity of computing its normalizing constant, which is among the most essential probabilistic inference tasks. Our complexity-theoretic results (almost) rule out the existence of efficient algorithms for this task unless the input matrices are forced to have favorable structures. In particular, we prove the following: (1) Computing $\sum_S\det({\bf A}_{S,S})^p$ exactly for every (fixed) positive even integer $p$ is UP-hard and Mod$_3$P-hard, which gives a negative answer to an open question posed by Kulesza and Taskar. (2) $\sum_S\det({\bf A}_{S,S})\det({\bf B}_{S,S})\det({\bf C}_{S,S})$ is NP-hard to approximate within a factor of $2^{O(|I|^{1-ε})}$ or $2^{O(n^{1/ε})}$ for any $ε>0$, where $|I|$ is the input size and $n$ is the order of the input matrix. This result is stronger than the #P-hardness for the case of two matrices derived by Gillenwater. (3) There exists a $k^{O(k)}n^{O(1)}$-time algorithm for computing $\sum_S\det({\bf A}_{S,S})\det({\bf B}_{S,S})$, where $k$ is the maximum rank of $\bf A$ and $\bf B$ or the treewidth of the graph formed by nonzero entries of $\bf A$ and $\bf B$. Such parameterized algorithms are said to be fixed-parameter tractable. These results can be extended to the fixed-size case. Further, we present two applications of fixed-parameter tractable algorithms given a matrix $\bf A$ of treewidth $w$: (4) We can compute a $2^{\frac{n}{2p-1}}$-approximation to $\sum_S\det({\bf A}_{S,S})^p$ for any fractional number $p>1$ in $w^{O(wp)}n^{O(1)}$ time. (5) We can find a $2^{\sqrt n}$-approximation to unconstrained MAP inference in $w^{O(w\sqrt n)}n^{O(1)}$ time.

cs.LG↗

arXiv2021

Reconfiguration Problems on Submodular Functions

Reconfiguration problems require finding a step-by-step transformation between a pair of feasible solutions for a particular problem. The primary concern in Theoretical Computer Science has been revealing their computational complexity for classical problems. This paper presents an initial study on reconfiguration problems derived from a submodular function, which has more of a flavor of Data Mining. Our submodular reconfiguration problems request to find a solution sequence connecting two input solutions such that each solution has an objective value above a threshold in a submodular function $f: 2^{[n]} \to \mathbb{R}_+$ and is obtained from the previous one by applying a simple transformation rule. We formulate three reconfiguration problems: Monotone Submodular Reconfiguration (MSReco), which applies to influence maximization, and two versions of Unconstrained Submodular Reconfiguration (USReco), which apply to determinantal point processes. Our contributions are summarized as follows: 1. We prove that MSReco and USReco are both $\mathsf{PSPACE}$-complete. 2. We design a $\frac{1}{2}$-approximation algorithm for MSReco and a $\frac{1}{n}$-approximation algorithm for (one version of) USReco. 3. We devise inapproximability results that approximating the optimum value of MSReco within a $(1-\frac{1+ε}{n^2})$-factor is $\mathsf{PSPACE}$-hard, and we cannot find a $(\frac{5}{6}+ε)$-approximation for USReco. 4. We conduct numerical study on the reconfiguration version of influence maximization and determinantal point processes using real-world social network and movie rating data.

cs.DS↗

arXiv2021

A Fully Polynomial Parameterized Algorithm for Counting the Number of Reachable Vertices in a Digraph

We consider the problem of counting the number of vertices reachable from each vertex in a digraph $G$, which is equal to computing all the out-degrees of the transitive closure of $G$. The current (theoretically) fastest algorithms run in quadratic time; however, Borassi has shown that this probl m is not solvable in truly subquadratic time unless the Strong Exponential Time Hypothesis fails [Inf. Process. Lett., 116(10):628--630, 2016]. In this paper, we present an $\mathcal{O}(f^3n)$-time exact algorithm, where $n$ is the number of vertices in $G$ and $f$ is the feedback edge number of $G$. Our algorithm thus runs in truly subquadratic time for digraphs of $f=\mathcal{O}(n^{\frac{1}{3}-ε})$ for any $ε> 0$, i.e., the number of edges is $n$ plus $\mathcal{O}(n^{\frac{1}{3}-ε})$, and is fully polynomial fixed parameter tractable, the notion of which was first introduced by Fomin, Lokshtanov, Pilipczuk, Saurabh, and Wrochna [ACM Trans. Algorithms, 14(3):34:1--34:45, 2018]. We also show that the same result holds for vertex-weighted digraphs, where the task is to compute the total weights of vertices reachable from each vertex.

cs.DS↗

arXiv2021

Spanning Tree Constrained Determinantal Point Processes are Hard to (Approximately) Evaluate

We consider determinantal point processes (DPPs) constrained by spanning trees. Given a graph $G=(V,E)$ and a positive semi-definite matrix $\mathbf{A}$ indexed by $E$, a spanning-tree DPP defines a distribution such that we draw $S\subseteq E$ with probability proportional to $\det(\mathbf{A}_S)$ only if $S$ induces a spanning tree. We prove $\sharp\textsf{P}$-hardness of computing the normalizing constant for spanning-tree DPPs and provide an approximation-preserving reduction from the mixed discriminant, for which FPRAS is not known. We show similar results for DPPs constrained by forests.

cs.LG↗

arXiv2021

Predictive Optimization with Zero-Shot Domain Adaptation

Prediction in a new domain without any training sample, called zero-shot domain adaptation (ZSDA), is an important task in domain adaptation. While prediction in a new domain has gained much attention in recent years, in this paper, we investigate another potential of ZSDA. Specifically, instead of predicting responses in a new domain, we find a description of a new domain given a prediction. The task is regarded as predictive optimization, but existing predictive optimization methods have not been extended to handling multiple domains. We propose a simple framework for predictive optimization with ZSDA and analyze the condition in which the optimization problem becomes convex optimization. We also discuss how to handle the interaction of characteristics of a domain in predictive optimization. Through numerical experiments, we demonstrate the potential usefulness of our proposed framework.

cs.LG↗

arXiv2020

The Solution Distribution of Influence Maximization: A High-level Experimental Study on Three Algorithmic Approaches

Influence maximization is among the most fundamental algorithmic problems in social influence analysis. Over the last decade, a great effort has been devoted to developing efficient algorithms for influence maximization, so that identifying the ``best'' algorithm has become a demanding task. In SIGMOD'17, Arora, Galhotra, and Ranu reported benchmark results on eleven existing algorithms and demonstrated that there is no single state-of-the-art offering the best trade-off between computational efficiency and solution quality. In this paper, we report a high-level experimental study on three well-established algorithmic approaches for influence maximization, referred to as Oneshot, Snapshot, and Reverse Influence Sampling (RIS). Different from Arora et al., our experimental methodology is so designed that we examine the distribution of random solutions, characterize the relation between the sample number and the actual solution quality, and avoid implementation dependencies. Our main findings are as follows: 1. For a sufficiently large sample number, we obtain a unique solution regardless of algorithms. 2. The average solution quality of Oneshot, Snapshot, and RIS improves at the same rate up to scaling of sample number. 3. Oneshot requires more samples than Snapshot, and Snapshot requires fewer but larger samples than RIS. We discuss the time efficiency when conditioning Oneshot, Snapshot, and RIS to be of identical accuracy. Our conclusion is that Oneshot is suitable only if the size of available memory is limited, and RIS is more efficient than Snapshot for large networks; Snapshot is preferable for small, low-probability networks.

cs.SI↗

arXiv2017

On the Power of Tree-Depth for Fully Polynomial FPT Algorithms

There are many classical problems in P whose time complexities have not been improved over the past decades. Recent studies of "Hardness in P" have revealed that, for several of such problems, the current fastest algorithm is the best possible under some complexity assumptions. To bypass this difficulty, Fomin et al. (SODA 2017) introduced the concept of fully polynomial FPT algorithms. For a problem with the current best time complexity $O(n^c)$, the goal is to design an algorithm running in $k^{O(1)}n^{c'}$ time for a parameter $k$ and a constant $c'<c$. In this paper, we investigate the complexity of graph problems in P parameterized by tree-depth, a graph parameter related to tree-width. We show that a simple divide-and-conquer method can solve many graph problems, including Weighted Matching, Negative Cycle Detection, Minimum Weight Cycle, Replacement Paths, and 2-hop Cover, in $O(\mathrm{td}\cdot m)$ time or $O(\mathrm{td}\cdot (m+n\log n))$ time, where $\mathrm{td}$ is the tree-depth of the input graph. Because any graph of tree-width $\mathrm{tw}$ has tree-depth at most $(\mathrm{tw}+1)\log_2 n$, our algorithms also run in $O(\mathrm{tw}\cdot m\log n)$ time or $O(\mathrm{tw}\cdot (m+n\log n)\log n)$ time. These results match or improve the previous best algorithms parameterized by tree-width. Especially, we solve an open problem of fully polynomial FPT algorithm for Weighted Matching parameterized by tree-width posed by Fomin et al.

cs.DS↗