Heterogeneity of agents in agent-based models (ABMs) is generally expressed as parameters of the agents that represent individuals in a population. The heterogeneity of agents is important when analyzing the effect of intervention measures on agents in ABMs. For realistic analysis, it is necessary to estimate agent heterogeneity from empirical data, but there are few studies on estimating agent heterogeneity in ABMs. In this study, we propose a method for estimating agent heterogeneity in ABMs. The proposed method estimates predictive distribution of the agent parameters using a particle filter and a variational Bayesian inference for aggregated observed data obtained from the real world. In order to evaluate the proposed method, we carry out a twin experiment and an empirical analysis using an ABM that represents infectious disease spread. Results of the twin experiment indicate that estimation of predictive distribution becomes stable as the number of particles of the particle filter increases. The results also show that the configuration of prior distribution for hierarchical parameters is important to properly extract the number of the clusters that express heterogeneity. Results of the empirical analysis demonstrate that the proposed method can estimate agent heterogeneity using observed data aggregated from real world population. We find that the proposed method has an advantage over methods based on detailed surveys of individuals in that it estimates the heterogeneity of individuals in a population using easily obtained aggregated observed data.

Citation count prediction is the task of predicting the future citation counts of academic papers, which is particularly useful for estimating the future impacts of an ever-growing number of academic papers. Although there have been many studies on citation count prediction, they are not applicable to predicting the citation counts of newly published papers, because they assume the availability of future citation counts for papers that have not had enough time passed since publication. In this paper, we first identify problems in the settings of existing studies and introduce a realistic citation count prediction task that strictly uses information available at the time of a target paper’s publication. For realistic citation prediction, we then propose two methods that leverage the citation counts of papers shortly after publication to capture the research trend that is important for predicting the citation counts of newly published papers. Through experiments using papers collected from arXiv and bioRxiv, we demonstrate that our methods considerably improve the performance of citation count prediction for newly published papers in a realistic setting.

The development of the COVID-19 vaccine and vaccination campaign was a significant concern for the people. In Japan, mass vaccination was initiated later than in other countries such as the United States, China, and Europe; however, vaccination coverage increased rapidly in this country, and in October 2021, Japan ranked 14th out of 229 countries in terms of COVID-19 vaccination rates. How did public opinion and concerns evolve in the face of the uncertain COVID-19 vaccination period? To address this question, we collected over 100 million Japanese vaccine-related tweets from January 1 to October 31, 2021. Using the Latent Dirichlet Allocation (LDA) model, we identified 15 main topics from a subset of tweets. We manually grouped these topics into four themes based on typical tweet content: (1) personal issues, (2) breaking news, (3) politics, and (4) conspiracy and humor. Then, we constructed hypotheses about topic evolution by interpreting the narrative underlying the tweets. We carefully read approximately 15,000 representative tweets and the percentage of a word in each topic to interpret the narrative. Finally, we verified the hypotheses by visualizing the change in the percentage of a word during the vaccination period. There are three main findings in this paper. First, the percentage of tweets containing “fear” and “anxiety” was highest in January 2021 and then decreased. This finding suggests that Twitter users felt fear and anxiety in January, when the vaccination schedule was unclear and that their negative feelings subsided once vaccination began. Second, the Twitter discourse reflected changes in the target population for vaccination, transitioning from discussions about health care workers in February to older individuals in April, and later to general Twitter users after July. Third, as the vaccination process progressed, users increasingly shared their real-time experiences through the tweets.

The proper operation of security guards is one of the major issues in public institutions. By assigning the security guard to the most appropriate duty areas and time-slots, the service can be provided in the best conceivable way. Security Guard Scheduling Problem (SGSP) is a combinatorial optimization problem, in which a set of security guards must be assigned into a limited set of working slots, subject to a given set of constraints. In the real world, it is natural to consider the scheduled security guard’s unexpected absences, e.g., illness, accident and injury of a security guard. In addition, it is required to respond immediately to the unexpected additional security works in case of emergency. Robustness, i.e., the ability to satisfy the security level even if some security guards break down, and the ability to respond immediately to the unexpected additional security works, is an expected property of a security guard team. In this paper, the main focus is laid on the Robust Security Guard Scheduling Problem (R-SGSP) . Two formal frameworks are defined, namely Proactive Security Guard Scheduling Problem with Robustness (R-SGSPpro) and Reactive Security Guard Scheduling Problem with Robustness (R-SGSPre), and some decision and optimization problems for R-SGSPpro and R-SGSPre are provided. Furthermore, zero-one integer programming formulations for the R-SGSPpro and R-SGSPre are provided. Finally, some empirical results for R-SGSPpro and R-SGSPre are reported.

These days the demand for non-intrusive occupancy detection using low time frequency e.g. 30 minutes household energy data from smart meters has been increased, because it is beneficial to society in many applicable areas such as removing absent delivery and controlling supply of energy automatically. There are privacy concerns for the public thus gathering the supervised occupancy labels on a large scale is infeasible. Previous signal processing studies in unsupervised domain adaptation realise unsupervised prediction, though assumption that a distribution shifts once by domains is inappropriate for non-intrusive occupancy detection because two time shifts occur naturally between seasons and between households. This paper proposes novel unsupervised domain adaptation strategy interseasons and inter-households domain adversarial neural networks (isih-DA) using domain adversarial neural networks (DANNs) and pseudo labeling, which smoothly learns distribution gaps in two dimensional domains. The isih-DA splits the problem into two sub-problems converting domain adaptation between seasons and between households into (1)domain adaptation between seasons (2)domain adaptation between households, thus divergence of each data pattern in sub-problems is decreased compared with direct domain adaptation between seasons and between households, finally this eases learning convergence in terms of classification performance for target data. We demonstrated experimental superiority over the previous studies using publicly largest ECO data set and temporal shift synthetic data we created via ECO data set. Temporal shift synthetic data corresponds to a simplified version of the situation inherent in the data that isih-DA excels at modeling.

Supervised learning, especially supervised deep learning, requires large amounts of labeled data. One approach to collecting large amounts of labeled data is by using a crowdsourcing platform where numerous workers perform the annotation tasks. However, the annotation results often contain label noise, as the annotation skills vary depending on the crowd workers and their ability to complete the task correctly. Learning from Crowds is a framework that directly trains the models using noisy labeled data from crowd workers. In this study, we propose a novel Learning from Crowds model inspired by SelectiveNet proposed for the selective prediction problem. The proposed method, the Label Selection Layer, trains a prediction model by automatically determining whether to use a worker ’s label for training using a selector network. A major advantage of the proposed method is that it can be applied to almost all variants of supervised learning problems by simply adding a selector network and changing the objective function for existing models without explicitly assuming a model of the noise in crowd annotations. The experimental results show that the performance of the proposed method is almost equivalent to or better than the Crowd Layer, which is one of the state-of-the-art methods for Deep Learning from Crowds, except for the regression problem case.

We propose a model for predicting the probability distribution of score in Self-Play deep reinforcement learning, which is used in game AI. In the proposed model, the probability distribution of score is obtained instead of expected value of score that is commonly used. By using it directly, the performance degradation problem in score learning is solved. Evaluation experiments comparing the proposed model with existing models show that the performance degradation problem is solved. Furthermore, the proposed model allowed more precise manipulation of score.