Predictive modeling competitions provide a new data mining approach that leverages crowds of data scientists to examine a wide variety of predictive models and build the best performance model. Competition hosts, who provide their own dataset and specify the problem to be solved, are not only able to obtain the best model from among those submitted but also to aggregate the submitted models to obtain one that outperforms the rest. In this paper, we report the results of a study conducted on CrowdSolving, a platform for predictive modeling competitions in Japan. We hosted a competition on a link prediction task and observed that (i) the prediction performance of the winner significantly outperformed that of a state-of-the-art method, (ii) the aggregated model constructed from all submitted models further improved the final performance, and (iii) the performance of the aggregated model built only from early submissions nevertheless overtook the final performance of the winner. Our results show the power of crowds for predictive modeling, not only in the quality of the obtained model, but also in its speed to achieve it. Furthermore, they demonstrate the possibilities of combining human insights and machine learning in data analytics.

Predictive modeling competitions provide a new data mining approach that leverages crowds of data scientists to examine a wide variety of predictive models and build the best performance model. In this paper, we report the results of a study conducted on CrowdSolving, a platform for predictive modeling competitions in Japan. We hosted a competition on a link prediction task and observed that (i) the prediction performance of the winner significantly outperformed that of a state-of-the-art method, (ii) the aggregated model constructed from all submitted models further improved the final performance, and (iii) the performance of the aggregated model built only from early submissions nevertheless overtook the final performance of the winner.

In this report, we designed a procedure for exploring the cellular relationships from the sets of characteristic genes in distinctive cell types, by using partial canonical correlation analysis. We applied the present procedure to the characteristic gene sets of seven subtypes of testicular germ cell tumors in a previous report. The cellular relationships were well reconstructed, consistent with the general histologic lineage model, and new implications for the subtype differentiation were found. In particular, the correspondence between the inferred relationships and the functional characterizations of constituent genes suggests a hypothesis for the classification between seminoma and embryonal carcinoma. The partial canonical correlation analysis is also appropriate for revealing new features of cellular relationships, based on the transcriptional programs. Thus, the present procedure helps to create a macroscopic view of the cellular relationships, by following the detection of the characteristically expressed genes.

We developed a procedure for indentifying transcriptional master regulators (MRs) related to special biological phenomena, such as diseases, in conjunction with network screening and inference. Network screening is a system for detecting activated transcriptional regulatory networks under particular conditions, based on the estimation of the graph structure consistency with the measured data. Since the network screening utilizes the known transcriptional factor (TF)-gene relationships as the experimental evidence for the molecular relationships, its performance depends on the ensemble of known TF networks used for its analysis. To compensate for its restrictions, a network inference method, the path consistency algorithm, is concomitantly utilized to identify MRs. The performance is illustrated by means of the known MRs in brain tumors that were computationally inferred and experimentally verified. As a result, the present procedure worked well for identifying MRs, in comparison to the previous computational selection for experimental verification.

The assessment of co-expressed genes is an initial step to investigate the gene regulatory system in a microarray analysis. We designed a simple method to assess co-expressed genes from the expression profiles, based on the path consistency (PC) algorithm, which systematically infers a causal graph. The PC algorithm is simply modified for the frequent occurrence of high correlations between expression profiles, which causes an accidental stop of the algorithm by the violation of a numerical calculation. By utilizing the flexible features of the algorithm, biological information about co-expressed genes can be introduced into the algorithm. The performance is illustrated by the application to the gene profiles of operons: 1168 gene profiles in 137 known operons in Escherichia coli. By the improved algorithm, more than 80% of the operons were correctly detected, and less than one gene per operon was falsely detected. The performance promises the wide feasibility of the present method for the assessment of co-expressed genes from microarray data on a genomic scale.

Because marker-based position tracking system is inexpensive and easy to use, it has the potential to make the system more feasible for homes or businesses and broaden the current suite of augmented reality (AR) techniques. To apply marker-based systems to homes or businesses, blending markers with environments naturally is important. Our innovative approach to marker-based 3D position tracking uses seamless patterns encrypted with positional data. Although users can obtain 3D positional data by processing marker images similar to many existing marker-based systems, our markers are designed with interior decoration in mind. That way they can be installed in walls, floors, or ceilings. Unlike existing systems whose fiducial markers were designed first and foremost to be processed by computers, ours are visually attractive. By integrating positional information within the interior design, our system enables users to enjoy the benefits of position tracking without being constantly aware of the system's presence. We developed a method for making patterns in which positional information is encrypted and a method for calculating the 3D position of a user by decoding those patterns. We then constructed a system using three patterns that were made by the proposed method and evaluated that system

Recent development in mobile and ubiquitous computing technology with broadband wireless network has realizing the location based application for mobile and wearable computers. According to the wide spreading number of cellular phone with GPS receiver, it has become easy to provide location based application in outdoor scene. On the contrary, there are several positioning technologies for indoor scene, such using wireless sensors or image processing technology. However, those technologies are still not utilized in daily environment due to the cost inefficiency and the complexity in system setup. Considering the problems in previous positioning system, it is necessary to overcome the following four features to utilize the system in daily scene.

UNLABELLED: We constructed a website for inferring a network by applying the graphical Gaussian model, from a large amount of data, including redundant information. The available tools on the website are based on a system, named ASIAN (Automatic System for Inferring A Network), in combination with the two methods in our previous papers, which were designed to analyze gene expression profiles on a genomic scale. One of the remarkable features of the website is its ability to infer a network, concomitant with hierarchical clustering and the following estimation of cluster boundaries. AVAILABILITY: http://eureka.ims.u-tokyo.ac.jp/asian

This paper describes a routing algorithm that assigns each arriving packet to one of multiple homogeneous parallel servers. The algorithm differs from existing routing policies in that it considers the impact of each arriving packet's size on the average packet delay. The algorithm is based on mimic optimal routing, which was derived from numerical solutions to a packet assignment problem. It works under the condition that the inter-arrival time and packet-size distributions are unknown, but a distribution set to which these two distributions belong is known in advance. The algorithm calculates statistics, such as means of these distributions, from input traffic, so it works under dynamic input traffic, where such distributions vary with time.