arXiv subjects

Kohei Hayashi

Publications and source records attributed to Kohei Hayashi.

103 records · Page 6Linked to original sources

arXiv2016

Minimizing Quadratic Functions in Constant Time

A sampling-based optimization method for quadratic functions is proposed. Our method approximately solves the following $n$-dimensional quadratic minimization problem in constant time, which is independent of $n$: $z^*=\min_{\mathbf{v} \in \mathbb{R}^n}\langle\mathbf{v}, A \mathbf{v}\rangle + n\langle\mathbf{v}, \mathrm{diag}(\mathbf{d})\mathbf{v}\rangle + n\langle\mathbf{b}, \mathbf{v}\rangle$, where $A \in \mathbb{R}^{n \times n}$ is a matrix and $\mathbf{d},\mathbf{b} \in \mathbb{R}^n$ are vectors. Our theoretical analysis specifies the number of samples $k(δ, ε)$ such that the approximated solution $z$ satisfies $|z - z^*| = O(εn^2)$ with probability $1-δ$. The empirical performance (accuracy and runtime) is positively confirmed by numerical experiments.

cs.LG↗

arXiv2016

Dark matter annihilation and decay from non-spherical dark halos in the Galactic dwarf satellites

The dwarf spheroidal galaxies (dSphs) in the Milky Way are the primary targets in the indirect searches for particle dark matter. To set robust constraints on candidate of dark matter particles, understanding the dark halo structure of these systems is of substantial importance. In this paper, we first evaluate the astrophysical factors for dark matter annihilation and decay for 24 dSphs, with taking into account a non-spherical dark halo, using generalized axisymmetric mass models based on axisymmetric Jeans equations. First, from a fitting analysis of the most recent kinematic data available, our axisymmetric mass models are a much better fit than previous spherical ones, thus, our work should be the most realistic and reliable estimator for astrophysical factors. Secondly, we find that among analysed dSphs, the ultra-faint dwarf galaxies Triangulum II and Ursa Major II are the most promising but large uncertain targets for dark matter annihilation while the classical dSph Draco is the most robust and detectable target for dark matter decay. It is also found that the non-sphericity of luminous and dark components influences the estimate of astrophysical factors, even though these factors largely depend on the sample size, the prior range of parameters and the spatial extent of the dark halo. Moreover, owing to these effects, the constraints on the dark matter annihilation cross-section are more conservative than those of previous spherical works. These results are important for optimizing and designing dark matter searches in current and future multi-messenger observations by space and ground-based telescopes.

astro-ph.GA↗

arXiv2016

Making Tree Ensembles Interpretable

Tree ensembles, such as random forest and boosted trees, are renowned for their high prediction performance, whereas their interpretability is critically limited. In this paper, we propose a post processing method that improves the model interpretability of tree ensembles. After learning a complex tree ensembles in a standard way, we approximate it by a simpler model that is interpretable for human. To obtain the simpler model, we derive the EM algorithm minimizing the KL divergence from the complex ensemble. A synthetic experiment showed that a complicated tree ensemble was approximated reasonably as interpretable.

stat.ML↗

arXiv2016

Doubly Decomposing Nonparametric Tensor Regression

Nonparametric extension of tensor regression is proposed. Nonlinearity in a high-dimensional tensor space is broken into simple local functions by incorporating low-rank tensor decomposition. Compared to naive nonparametric approaches, our formulation considerably improves the convergence rate of estimation while maintaining consistency with the same function class under specific conditions. To estimate local functions, we develop a Bayesian estimator with the Gaussian process prior. Experimental results show its theoretical properties and high performance in terms of predicting a summary statistic of a real complex network.

stat.ML↗

arXiv2016

A Tractable Fully Bayesian Method for the Stochastic Block Model

The stochastic block model (SBM) is a generative model revealing macroscopic structures in graphs. Bayesian methods are used for (i) cluster assignment inference and (ii) model selection for the number of clusters. In this paper, we study the behavior of Bayesian inference in the SBM in the large sample limit. Combining variational approximation and Laplace's method, a consistent criterion of the fully marginalized log-likelihood is established. Based on that, we derive a tractable algorithm that solves tasks (i) and (ii) concurrently, obviating the need for an outer loop to check all model candidates. Our empirical and theoretical results demonstrate that our method is scalable in computation, accurate in approximation, and concise in model selection.

cs.LG↗

arXiv2015

Bayesian Masking: Sparse Bayesian Estimation with Weaker Shrinkage Bias

A common strategy for sparse linear regression is to introduce regularization, which eliminates irrelevant features by letting the corresponding weights be zeros. However, regularization often shrinks the estimator for relevant features, which leads to incorrect feature selection. Motivated by the above-mentioned issue, we propose Bayesian masking (BM), a sparse estimation method which imposes no regularization on the weights. The key concept of BM is to introduce binary latent variables that randomly mask features. Estimating the masking rates determines the relevance of the features automatically. We derive a variational Bayesian inference algorithm that maximizes the lower bound of the factorized information criterion (FIC), which is a recently developed asymptotic criterion for evaluating the marginal log-likelihood. In addition, we propose reparametrization to accelerate the convergence of the derived algorithm. Finally, we show that BM outperforms Lasso and automatic relevance determination (ARD) in terms of the sparsity-shrinkage trade-off.

stat.ML↗

arXiv2015

Structural properties of non-spherical dark halos in Milky Way and Andromeda dwarf spheroidal galaxies

We investigate the non-spherical density structure of dark halos of the dwarf spheroidal (dSph) galaxies in the Milky Way and Andromeda galaxies based on revised axisymmetric mass models from our previous work. The models we adopt here fully take into account velocity anisotropy of tracer stars confined within a flattened dark halo. Applying our models to the available kinematic data of the 12 bright dSphs, we find that these galaxies associate with, in general, elongated dark halos, even considering the effect of this velocity anisotropy of stars. We also find that the best-fit parameters, especially for the shapes of dark halos and velocity anisotropy, are susceptible to both the availability of velocity data in the outer regions and the effect of the lack of sample stars in each spatial bin. Thus, to obtain more realistic limits on dark halo structures, we require photometric and kinematic data over much larger areas in the dSphs than previously explored. The results obtained from the currently available data suggest that the shapes of dark halos in the dSphs are more elongated than those of $Λ$CDM subhalos. This mismatch needs to be solved by theory including baryon components and the associated feedback to dark halos as well as by further observational limits in larger areas of dSphs. It is also found that more diffuse dark halos may have undergone consecutive star formation history, thereby implying that dark-halo structure plays an important role in star formation activity.

astro-ph.GA↗

arXiv2015

Rebuilding Factorized Information Criterion: Asymptotically Accurate Marginal Likelihood

Factorized information criterion (FIC) is a recently developed approximation technique for the marginal log-likelihood, which provides an automatic model selection framework for a few latent variable models (LVMs) with tractable inference algorithms. This paper reconsiders FIC and fills theoretical gaps of previous FIC studies. First, we reveal the core idea of FIC that allows generalization for a broader class of LVMs, including continuous LVMs, in contrast to previous FICs, which are applicable only to binary LVMs. Second, we investigate the model selection mechanism of the generalized FIC. Our analysis provides a formal justification of FIC as a model selection criterion for LVMs and also a systematic procedure for pruning redundant latent variables that have been removed heuristically in previous studies. Third, we provide an interpretation of FIC as a variational free energy and uncover a few previously-unknown their relationships. A demonstrative study on Bayesian principal component analysis is provided and numerical experiments support our theoretical results.

cs.LG↗

arXiv2015

A common surface-density scale for the Milky Way and Andromeda dwarf satellites as a constraint on dark matter models

In an attempt to place an explicit constraint on dark matter models, we define and estimate a mean surface density of a dark halo within a radius of maximum circular velocity, which is derivable for various galaxies with any dark-matter density profiles. We find that this surface density is generally constant across a wide range of maximum circular velocities of $\sim$ 10 to 400 km s$^{-1}$, irrespective of different density distribution in each of the galaxies. This common surface density at high halo-mass scales is found to be naturally reproduced by both cold and warm dark matter (CDM and WDM) models, even without employing any fitting procedures. However, the common surface density at dwarf-galaxy scales, for which we have derived from the Milky Way and Andromeda dwarf satellites, is reproduced only in a massive range of WDM particle masses, whereas CDM provides a reasonable agreement with the observed constancy. This is due to the striking difference between mass-concentration relations for CDM and WDM halos at low halo-mass scales. In order to explain the universal surface density of dwarf-galaxy scales in WDM models, we suggest that WDM particles need to be heavier than 3 keV.

astro-ph.GA↗

arXiv2014

The prolate dark matter halo of the Andromeda galaxy

We present new limits on the global shape of the dark matter halo in the Andromeda galaxy using and generalizing non-spherical mass models developed by Hayashi & Chiba and compare our results with theoretical predictions of cold dark matter (CDM) models. This is motivated by the fact that CDM models predict non-spherical virialized dark halos, which reflect the process of mass assembly in the galactic scale. Applying our models to the latest kinematic data of globular clusters and dwarf spheroidal galaxies in the Andromeda halo, we find that the most plausible cases for Andromeda yield a prolate shape for its dark halo, irrespective of assumed density profiles. We also find that this prolate dark halo in Andromeda is consistent with theoretical predictions in which the satellites are distributed anisotropically and preferentially located along major axes of their host halos. It is a reflection of the intimate connection between galactic dark matter halos and the cosmic web. Therefore, our result is profound in understanding internal dynamics of halo tracers in Andromeda, such as orbital evolutions of tidal stellar streams, which play important roles in extracting the abundance of CDM subhalos through their dynamical effects on stream structures.

astro-ph.GA↗

arXiv2012

Factorized Asymptotic Bayesian Hidden Markov Models

This paper addresses the issue of model selection for hidden Markov models (HMMs). We generalize factorized asymptotic Bayesian inference (FAB), which has been recently developed for model selection on independent hidden variables (i.e., mixture models), for time-dependent hidden variables. As with FAB in mixture models, FAB for HMMs is derived as an iterative lower bound maximization algorithm of a factorized information criterion (FIC). It inherits, from FAB for mixture models, several desirable properties for learning HMMs, such as asymptotic consistency of FIC with marginal log-likelihood, a shrinkage effect for hidden state selection, monotonic increase of the lower FIC bound through the iterative optimization. Further, it does not have a tunable hyper-parameter, and thus its model selection process can be fully automated. Experimental results shows that FAB outperforms states-of-the-art variational Bayesian HMM and non-parametric Bayesian HMM in terms of model selection accuracy and computational efficiency.

cs.LG↗

arXiv2012

Probing non-spherical dark halos in the Galactic dwarf galaxies

We construct axisymmetric mass models for dwarf spheroidal (dSph) galaxies in the Milky Way to obtain plausible limits on the non-spherical structure of their dark halos. This is motivated by the fact that the observed luminous parts of the dSphs are actually non-spherical and Cold Dark Matter (CDM) models predict non-spherical virialized dark halos. Our models consider velocity anisotropy of stars $\bar{v^2_R} / \bar{v^2_ϕ}$, which can vary with the adopted cylindrical coordinates under the assumption $\bar{v^2_z}=\bar{v^2_R}$ for simplicity, and also include an inclination of the system as a fitting parameter to explain the observed line-of-sight velocity dispersion profile. Applying these models to six of the bright dSphs in the Milky Way, we find that the best-fitting cases for most of the dSphs yield oblate and flattened dark halos, irrespective of assumed density profiles in their central parts. We also find that the total mass of the dSphs enclosed within a spheroid with major-axis length of 300 pc varies from $10^6M_{\odot}$ to $10^7M_{\odot}$, contrary to the conclusion from spherical models. This suggests the importance of considering shapes of dark halos in mass models of the dSphs. It is also found that dark halos of the Galactic dSphs may be more flattened than N-body predictions, thereby implying our yet incomplete understanding of baryonic and/or non-baryonic dark matter physics in dwarf galaxy scales.

astro-ph.CO↗

arXiv2011

Estimation of low-rank tensors via convex optimization

In this paper, we propose three approaches for the estimation of the Tucker decomposition of multi-way arrays (tensors) from partial observations. All approaches are formulated as convex minimization problems. Therefore, the minimum is guaranteed to be unique. The proposed approaches can automatically estimate the number of factors (rank) through the optimization. Thus, there is no need to specify the rank beforehand. The key technique we employ is the trace norm regularization, which is a popular approach for the estimation of low-rank matrices. In addition, we propose a simple heuristic to improve the interpretability of the obtained factorization. The advantages and disadvantages of three proposed approaches are demonstrated through numerical experiments on both synthetic and real world datasets. We show that the proposed convex optimization based approaches are more accurate in predictive performance, faster, and more reliable in recovering a known multilinear structure than conventional approaches.

stat.ML↗