arXiv subjects

Koji Tsuda

Publications and source records attributed to Koji Tsuda.

48 records · Page 3Linked to original sources

arXiv2017

Machine learning reveals orbital interaction in crystalline materials

We propose a novel representation of crystalline materials named orbital-field matrix (OFM) based on the distribution of valence shell electrons. We demonstrate that this new representation can be highly useful in mining material data. Our experiment shows that the formation energies of crystalline materials, the atomization energies of molecular materials, and the local magnetic moments of the constituent atoms in transition metal--rare-earth metal bimetal alloys can be predicted with high accuracy using the OFM. Knowledge regarding the role of coordination numbers of transition-metal and rare-earth metal elements in determining the local magnetic moment of transition metal sites can be acquired directly from decision tree regression analyses using the OFM.

cond-mat.mtrl-sci↗

arXiv2016

Designing nanostructures for interfacial phonon transport via Bayesian optimization

We demonstrate optimization of thermal conductance across nanostructures by developing a method combining atomistic Green's function and Bayesian optimization. With an aim to minimize and maximize the interfacial thermal conductance (ITC) across Si-Si and Si-Ge interfaces by means of Si/Ge composite interfacial structure, the method identifies the optimal structures from calculations of only a few percent of the entire candidates (over 60,000 structures). The obtained optimal interfacial structures are non-intuitive and impacting: the minimum-ITC structure is an aperiodic superlattice that realizes 50% reduction from the best periodic superlattice. The physical mechanism of the minimum ITC can be understood in terms of crossover of the two effects on phonon transport: as the layer thickness in superlattice increases, the impact of Fabry-Pérot interference increases, and the rate of reflection at the layer-interfaces decreases. Aperiodic superlattice with spatial variation in the layer thickness has a degree of freedom to realize optimal balance between the above two competing mechanism. Furthermore, aperiodicity breaks the constructive phonon interference between the interfaces inhibiting the coherent phonon transport. The present work shows the effectiveness and advantage of material informatics in designing nanostructures to control heat conduction, which can be extended to other interfacial structures.

cond-mat.mtrl-sci↗

arXiv2016

Information Decomposition on Structured Space

We build information geometry for a partially ordered set of variables and define the orthogonal decomposition of information theoretic quantities. The natural connection between information geometry and order theory leads to efficient decomposition algorithms. This generalization of Amari's seminal work on hierarchical decomposition of probability distributions on event combinations enables us to analyze high-order statistical interactions arising in neuroscience, biology, and machine learning.

cs.IT↗

arXiv2016

Selective Inference Approach for Statistically Sound Predictive Pattern Mining

Discovering statistically significant patterns from databases is an important challenging problem. The main obstacle of this problem is in the difficulty of taking into account the selection bias, i.e., the bias arising from the fact that patterns are selected from extremely large number of candidates in databases. In this paper, we introduce a new approach for predictive pattern mining problems that can address the selection bias issue. Our approach is built on a recently popularized statistical inference framework called selective inference. In selective inference, statistical inferences (such as statistical hypothesis testing) are conducted based on sampling distributions conditional on a selection event. If the selection event is characterized in a tractable way, statistical inferences can be made without minding selection bias issue. However, in pattern mining problems, it is difficult to characterize the entire selection process of mining algorithms. Our main contribution in this paper is to solve this challenging problem for a class of predictive pattern mining problems by introducing a novel algorithmic framework. We demonstrate that our approach is useful for finding statistically significant patterns from databases.

stat.ML↗

arXiv2016

Safe Pattern Pruning: An Efficient Approach for Predictive Pattern Mining

In this paper we study predictive pattern mining problems where the goal is to construct a predictive model based on a subset of predictive patterns in the database. Our main contribution is to introduce a novel method called safe pattern pruning (SPP) for a class of predictive pattern mining problems. The SPP method allows us to efficiently find a superset of all the predictive patterns in the database that are needed for the optimal predictive model. The advantage of the SPP method over existing boosting-type method is that the former can find the superset by a single search over the database, while the latter requires multiple searches. The SPP method is inspired by recent development of safe feature screening. In order to extend the idea of safe feature screening into predictive pattern mining, we derive a novel pruning rule called safe pattern pruning (SPP) rule that can be used for searching over the tree defined among patterns in the database. The SPP rule has a property that, if a node corresponding to a pattern in the database is pruned out by the SPP rule, then it is guaranteed that all the patterns corresponding to its descendant nodes are never needed for the optimal predictive model. We apply the SPP method to graph mining and item-set mining problems, and demonstrate its computational advantage.

stat.ML↗

arXiv2015

Redesigning pattern mining algorithms for supercomputers

Upcoming many core processors are expected to employ a distributed memory architecture similar to currently available supercomputers, but parallel pattern mining algorithms amenable to the architecture are not comprehensively studied. We present a novel closed pattern mining algorithm with a well-engineered communication protocol, and generalize it to find statistically significant patterns from personal genome data. For distributing communication evenly, it employs global load balancing with multiple stacks distributed on a set of cores organized as a hypercube with random edges. Our algorithm achieved up to 1175-fold speedup by using 1200 cores for solving a problem with 11,914 items and 697 transactions, while the naive approach of separating the search space failed completely.

cs.DC↗

arXiv2015

Safe Feature Pruning for Sparse High-Order Interaction Models

Taking into account high-order interactions among covariates is valuable in many practical regression problems. This is, however, computationally challenging task because the number of high-order interaction features to be considered would be extremely large unless the number of covariates is sufficiently small. In this paper, we propose a novel efficient algorithm for LASSO-based sparse learning of such high-order interaction models. Our basic strategy for reducing the number of features is to employ the idea of recently proposed safe feature screening (SFS) rule. An SFS rule has a property that, if a feature satisfies the rule, then the feature is guaranteed to be non-active in the LASSO solution, meaning that it can be safely screened-out prior to the LASSO training process. If a large number of features can be screened-out before training the LASSO, the computational cost and the memory requirment can be dramatically reduced. However, applying such an SFS rule to each of the extremely large number of high-order interaction features would be computationally infeasible. Our key idea for solving this computational issue is to exploit the underlying tree structure among high-order interaction features. Specifically, we introduce a pruning condition called safe feature pruning (SFP) rule which has a property that, if the rule is satisfied in a certain node of the tree, then all the high-order interaction features corresponding to its descendant nodes can be guaranteed to be non-active at the optimal solution. Our algorithm is extremely efficient, making it possible to work, e.g., with 3rd order interactions of 10,000 original covariates, where the number of possible high-order interaction features is greater than 10^{12}.

stat.ML↗

arXiv2015

Discovery of low thermal conductivity compounds with first-principles anharmonic lattice dynamics calculations and Bayesian optimization

Compounds of low lattice thermal conductivity (LTC) are essential for seeking thermoelectric materials with high conversion efficiency. Some strategies have been used to decrease LTC. However, such trials have yielded successes only within a limited exploration space. Here we report the virtual screening of a library containing 54,779 compounds. Our strategy is to search the library through Bayesian optimization using for the initial data the LTC obtained from first-principles anharmonic lattice dynamics calculations for a set of 101 compounds. We discovered 221 materials with very low LTC. Two of them have even an electronic band gap < 1 eV, what makes them exceptional candidates for thermoelectric applications. In addition to those newly discovered thermoelectric materials, the present strategy is believed to be powerful for many other applications in which chemistry of materials are required to be optimized.

cond-mat.mtrl-sci↗

arXiv2014

Machine learning with systematic density-functional theory calculations: Application to melting temperatures of single and binary component solids

A combination of systematic density functional theory (DFT) calculations and machine learning techniques has a wide range of potential applications. This study presents an application of the combination of systematic DFT calculations and regression techniques to the prediction of the melting temperature for single and binary compounds. Here we adopt the ordinary least-squares regression (OLSR), partial least-squares regression (PLSR), support vector regression (SVR) and Gaussian process regression (GPR). Among the four kinds of regression techniques, the SVR provides the best prediction. In addition, the inclusion of physical properties computed by the DFT calculation to a set of predictor variables makes the prediction better. Finally, a simulation to find the highest melting temperature toward the efficient materials design using kriging is demonstrated. The kriging design finds the compound with the highest melting temperature much faster than random designs. This result may stimulate the application of kriging to efficient materials design for a broad range of applications.

cond-mat.mtrl-sci↗

arXiv2011

LGM: Mining Frequent Subgraphs from Linear Graphs

A linear graph is a graph whose vertices are totally ordered. Biological and linguistic sequences with interactions among symbols are naturally represented as linear graphs. Examples include protein contact maps, RNA secondary structures and predicate-argument structures. Our algorithm, linear graph miner (LGM), leverages the vertex order for efficient enumeration of frequent subgraphs. Based on the reverse search principle, the pattern space is systematically traversed without expensive duplication checking. Disconnected subgraph patterns are particularly important in linear graphs due to their sequential nature. Unlike conventional graph mining algorithms detecting connected patterns only, LGM can detect disconnected patterns as well. The utility and efficiency of LGM are demonstrated in experiments on protein contact maps.

cs.DS↗

arXiv2009

Efficient Construction of Neighborhood Graphs by the Multiple Sorting Method

Neighborhood graphs are gaining popularity as a concise data representation in machine learning. However, naive graph construction by pairwise distance calculation takes $O(n^2)$ runtime for $n$ data points and this is prohibitively slow for millions of data points. For strings of equal length, the multiple sorting method (Uno, 2008) can construct an $ε$-neighbor graph in $O(n+m)$ time, where $m$ is the number of $ε$-neighbor pairs in the data. To introduce this remarkably efficient algorithm to continuous domains such as images, signals and texts, we employ a random projection method to convert vectors to strings. Theoretical results are presented to elucidate the trade-off between approximation quality and computation time. Empirical results show the efficiency of our method in comparison to fast nearest neighbor alternatives.

cs.DS↗

arXiv2002

Approximating Incomplete Kernel Matrices by the em Algorithm

In biological data, it is often the case that observed data are available only for a subset of samples. When a kernel matrix is derived from such data, we have to leave the entries for unavailable samples as missing. In this paper, we make use of a parametric model of kernel matrices, and estimate missing entries by fitting the model to existing entries. The parametric model is created as a set of spectral variants of a complete kernel matrix derived from another information source. For model fitting, we adopt the em algorithm based on the information geometry of positive definite matrices. We will report promising results on bacteria clustering experiments using two marker sequences: 16S and gyrB.

cs.LG↗