As data utilization in organizations is advancing in various fields, insights that data brings will be more diverse when it is sourced through collaboration across different organizations. Federated learning, a machine learning method with distributed data across organizations, with local differential privacy protects privacy by sharing only the model parameters and the information necessary for model update, without having to share the data each organization holds. However, there is a problem with local differential privacy, where the amount of noise increases, leading to the degradation in model accuracy. In this paper, we propose a method of reducing the impact of noise compared to conventional federated learning by leveraging private cross-organizational data collaboration, called Private Cross-aggregation Technology (PCT). PCT combines Private Set Intersection Cardinality, Trusted Execution Environment and Differential Privacy, and outputs a cross-tabulation table that is private from input to output. Our method consists of two steps: (1) creating a private cross-tabulation table using PCT, and (2) training a ML using the private cross-tabulation table. The experiment results showed that (1) the classification accuracy of the proposed method was higher than that of the baseline method in situations where the privacy budget is limited, and (2) the computation time of the proposed method was shorter than that of the baseline method.

In a disaster situation, local and municipal governments need to distribute relief supplies and provide administrative support to evacuee. Although people are supposed to evacuate to evacuation shelters designated by local governments, some people take refuge at non-designated facilities, called non-designated evacuation shelters, due to unavoidable circumstances such as damages on the access route to designated evacuation shelters. Upon occurrence of a disaster, therefore, it is necessary for the local governments to quickly find the locations of non-designated evacuation shelters. In this paper, we propose a method to detect non-designated evacuation shelters based on autoencoder (AE)-based anomaly detection using real-time population dynamics generated from operation data of cellular phone networks. We assume that reconstruction errors of an AE model include both the errors due to characteristic differences between locations and the errors due to anomalies in population dynamics. Thus, we propose to use the ratio of the reconstruction error before and after the earthquake to determine the threshold of anomaly detection. We evaluate the performance of the proposed method on data from three actual earthquakes in Japan. The evaluation results show that our reconstruction-error-based approach can achieve better accuracy for the actual disaster data compared to a baseline method that exploits statistical anomaly detection.

Relieving the growing traffic congestion is important not only to reduce damage to the environment but also to improve our quality of life. Since new construction and expansion of roads is expensive, existing roads should be utilized more effectively. One method for doing so is to disperse the traffic demands. To achieve this, it is necessary to provide highly accurate traffic jam prediction information before the drivers decide on their travel routes. However, it is somewhat difficult to accurately predict the traffic conditions several hours to half a day ahead when using traffic data. In this work, we propose a method for accurately predicting traffic conditions several hours to half a day ahead by exploiting real-time population data generated from cellular networks. Predicting non-linear traffic conditions from population data, which is high-dimensional and sparse data, is a non-linear regression problem in high-dimension, low-sample-size (HDLSS) situations. We therefore focus on traffic demand and assume there is a linear relationship between population and traffic demand. The proposed method predicts traffic demand from population (linear regression in HDLSS situations) and then travel time from the predicted traffic demand (nonlinear regression in non-HDLSS situations). The evaluation results showed that our method can predict the traffic condition several hours to half a day ahead with high accuracy, thus validating our hypothesis that there is a linear relationship between population and traffic demand. We also conducted a demonstration experiment on the Tokyo Wan Aqua-Line Expressway and the Kan-Etsu Expressway in which the prediction results by the proposed method were delivered to drivers. The responses to a questionnaire survey showed that behavior changes occurred in many of the drivers after they looked at the prediction result.

Storm and flood disasters such as typhoons and torrential rains are becoming more intense and frequent. The national government and municipalities must respond to such natural disasters as soon as possible. When the scale of damage is large; however, it takes much time to investigate the severity of damage, and the initial response can be delayed. If we could precisely and rapidly estimate the severity of damage for each city at an early stage, the national government would be able to better support the municipalities, and consequently respond quickly to help citizens. In this paper, we propose a novel approach to estimate the severity of disaster damage within a short time period after a disaster occurs by exploiting real-time population data generated from cellular networks. First, we investigate the relationship between real-time population data and the severity of damage. Then, we design a Graph Convolutional Networks for Disaster Damage Estimation, called D2E-GCN, which fully exploits the directed and weighted characteristics of human mobility graph. We conduct an offline evaluation on real-world datasets including two typhoons that hit Japan. The evaluation results show that the proposed method outperforms baseline methods which do not consider the graph structure of cities, and the proposed method can estimate the severity of damage approximately 48 hours after typhoons passed. Moreover, we find the experimental insight that the estimation performance can be significantly affected by the graph construction method for GCN models.

To utilize a huge amount of observation data based on real-world events, a data assimilation process is needed to estimate the state of the system behind the observed data. The Kalman filter is a very commonly used technique in data assimilation, but it has a problem in terms of practical use from the viewpoint of processing efficiency and estimating the deterioration in precision when applied to particularly large-scale datasets. In this paper, we propose a method that simultaneously addresses these problems and demonstrate its usefulness. The proposed method improves the processing efficiency and suppresses the deterioration in estimation precision by introducing correction processes focusing on the non-negative nature and sparseness of data in wavelet space. We show that the proposed method can accurately estimate population dynamics (MAE ≤ 3, RMSE ≤ 7) on the basis of an evaluation done using population data generated from cellular networks. In addition, the possibility of wide area abnormality detection using the proposed method is shown from a situation analysis of when Category 5 typhoon Hagibis made landfall in Japan. The proposed method has been deployed in a commercial service to estimate real-time population dynamics in Japan.