岩田 員典

Reducing traffic accidents is an important social issue, and traffic accident analysis based on vehicle behavior data has recently attracted attention. However, collecting such data continuously over a wide area is difficult because of measurement costs and privacy constraints. One possible solution is to use urban traffic simulations to reproduce traffic flow and vehicle behavior and obtain the data needed for analysis. However, applying urban traffic simulation to traffic accident analysis requires a framework to evaluate whether real-world traffic flow and vehicle behavior can be reproduced with sufficient accuracy. This study proposes such an evaluation framework comprising three perspectives: Traffic Flow Validity (TFV), Traffic Volume Similarity (TVS), and Vehicle Behavior Similarity (VBS). The framework was applied to an existing SUMO-based urban traffic simulation covering the Aichi Prefecture. The results showed that TFV was generally valid, whereas TVS and VBS differed from real-world conditions at many locations and time periods, indicating room for improvement. Overall, the proposed framework showed usefulness in evaluating the applicability of urban traffic simulations to traffic accident analysis.

GraphWaveNet, a Graph Neural Network (GNN) that can handle network structures contained in traffic datasets as inputs and outputs, has demonstrated high performance in traffic prediction. However, most existing studies evaluate models using highway datasets, and few focus on non-express roads. This study aims to clarify the prediction accuracy when applying GraphWaveNet to non-express roads and to reveal the graph structure influencing the prediction accuracy by analyzing the training data and learning weights of the GNN. The experimental results showed that the average prediction accuracy on non-express roads achieved a Mean Absolute Error (MAE) of 5.8136, Mean Absolute Percentage Error (MAPE) of 23.38%, Root Mean Square Error (RMSE) of 8.6613, and Weighted Absolute Percentage Error (WAPE) of 13.52%. Compared with the results from highway datasets, a similar accuracy was achieved in road groups with large traffic volumes. Furthermore, the factors extracted by factor analysis alongside the self-adaptive adjacency matrix suggested that GraphWaveNet learned the relationships between nodes that are not physically close together. As a future research direction, we aim to quantitatively evaluate the relationship between the extracted factor loadings and connection weights of the self-adaptive adjacency matrix using statistical methods.

The effectiveness of problem-solving using a Multi-agent System (MAS) depends on the design of cooperative strategies, but this design is difficult. The difficulty arises because the performance of cooperative strategies strongly depends on environmental characteristics. This study focusses on the RoboCupRescue Simulation, a disaster rescue simulation based on a multi-agent approach. To establish a map-adaptive strategy design method, we clarify how specific map characteristics affect individual rescue activities. We introduce new map and agent activity metrics, and analyze their relationship with cooperative strategy performance using LASSO regression and SHAP values. The results reveal that characteristics such as building density, D-value, and bridge ratios affect debris cleaning and movement efficiency. These findings clarify environmental dependency and provide a crucial foundation for designing map-adaptive strategies.

Traffic accidents remain a serious social problem, and effective prevention requires hotspot prediction of high-risk locations. Highly effective accident prediction must simultaneously satisfy two requirements: indicating the environmental factors that contribute to accident risks, and predicting at a microspatial resolution to identify target locations for countermeasures. Spatial-temporal network kernel density estimation (STNKDE), conventionally used for accident risk prediction, is robust against sparsity but does not consider environmental factors because it treats all accidents based solely on distance. Deep learning, by contrast, often degrades under extreme sparsity at microspatial resolutions. This study proposes PairEnv-STNKDE, a method that adopts role separation in which an ellipsoidal kernel structurally guarantees distance decay and a neural network learns environmental adjustments. By taking environmental features of the estimation and accident locations as paired inputs, the network adjusts each accident's contribution according to the environment. In an experiment in Naka Ward, Nagoya City, the proposed method achieved significantly higher prediction accuracy than conventional STNKDE. Furthermore, SHapley Additive exPlanations (SHAP) analysis confirmed contrasting effects between the estimation and accident locations, providing evidence to differentiate countermeasure directions based on the environment. This study contributes to realizing highly effective accident prediction that satisfies both requirements.

Recent major earthquakes in Japan have highlighted the growing importance of disaster preparedness. However, opportunities for household discussions on disaster preparedness remain insufficient, primarily due to a lack of effective prompts to initiate such conversations. To address this issues, this paper proposes a discussion-based evacuation route experience system using RoboCup Rescue Simulation (RRS) to encourage household discussions on disaster preparedness. The proposed system combines step-by-step questioning, evacuation route planning, and a multi-participant evacuation experience to help users recognize disaster preparedness as a personally relevant issue, thereby promoting discussion among participants. Content based on the proposed system was exhibited at an event, where a questionnaire survey was conducted among participants. The results confirmed that, in addition to stimulating discussions during the experience, the content effectively encouraged subsequent household conversations on disaster preparedness.

OLS-regression fails to provide meaningful solutions under large numbers of predictor variables due to the presence of multicollinearity. Sparse regression, or best subset selection, is used in such cases utilizing norm-0 control or norm-1 regularization. Mixed-integer optimization models resulting under norm-0 control, however, are computationally intractable although recent advances have been made for a moderate number of predictors. This paper contributes with a new efficient approach in very large dimensions under successive separable quadratic approximation of the mean squared error (MSE) function. At every iteration, given a current pivot solution, a separable form of the MSE function is minimized over a local hypercube trust region that is discretized to obtain an all-integer optimization subproblem employing norm-0 and norm-1 parametrization. Each subproblem is solved efficiently using the entropy-based constraint surrogation technique (ISCENT). The true MSE value associated with the subproblem optima is then used to specify a target MSE with specified tolerance, and the local trust region is enumerated to identify solutions that satisfy the target. With successively shrinking local hypercubes, along with corresponding subproblem optima and target enumeration, the method terminates with a high quality sparse predictive system. We test the method using two high-dimensional applications: financial index-tracking portfolio selection using 225 assets, and cancer prediction using genomic data having 906,600 predictors representing genetic variations for a sample of 704 humans. The proposed approach is shown to be more efficient and effective relative to the standard OLS or Lasso/Ridge models in providing accurate predictions.

交通事故はさまざまな要因から発生し，その状況は多くの項目からなる統計情報として記録されている．このデータは事故対策の立案のために，要因分析に利用されるが，その結果を対策へと結びつけるためには，分析の信頼性が重要である．近年では，膨大な交通事故データから潜在的な要因を抽出するために，機械学習を用いた分析がおこなわれている．本稿では，その事例の一つとして自己組織化マップを用いた要因分析を取り上げる．自己組織化マップのような教師なし学習では，収束性や解釈性から結果の信頼性を評価する必要がある．しかし，交通事故データは多くの項目を持ち，欠損やノイズも含む．そのため，学習の収束を得るには，特徴の選択や値の類型化など，適切な前処理が必要となる．また，収束が得られたとしても結果の解釈が難しい場合がある一方で，十分に収束していなくても有用な知見が得られることもある．機械学習を用いた交通事故分析では，このように入力データの複雑さに依存した，いくらかの課題がある．本稿では，自己組織化マップを用いた事例を通じ，機械学習を用いた交通事故の要因分析において，信頼性のある結果を得るための方法について解説する．

Machine learning, deep learning and neural networks are extensively developed in many fields, with neural networks playing an important role in a wide variety of applications. However, a sufficient explanation of the structure and functionality of complex and deep neural networks is still needed. In this paper, it is shown that bio-inspired networks are useful for the explanation of network functions. First, the asymmetric network is created based on the biological retinal networks. Second, the classification performance of the asymmetric network is compared to that of the symmetric networks. The directional vectors in the asymmetric networks are generated on the adjacent neurons caused by movement stimulus, which create independent subspaces. Vectors for the movement stimulus are reported experimentally to be generated in the layered cortex in the brain. In this paper, it is shown computationally that many directional movement vectors are generated in the layered asymmetric networks, which create also independent subspaces. Further, when the correlational activities of the adjacent cells are represented in the directed vectors, they create independent subspaces than the direct inputs in the networks. These asymmetric subnetworks will facilitate the transmission of sensory information to higher-level processes such as efficient feature extraction, classification, and learning in the layered networks.

本稿では英語の発音を改善するための方策として，英単語の発音を大きく 3 つに分類し，それぞれに合わせた練習方法を提案する．そして，その手法によりどの程度発音が改善されたのか検証を行う．さらに， 効果的な練習をするために， その方策に従った練習をするための Web アプリも作成する．本アプリでは， 様々な例文を LLM （Large Language Model：大規模言語モデル) の Few-shot learning により作成できるようにし， 例文に対応した合成音声も生成する．

PLATEAU（プラトー）とは、2020年より国土交通省が推進している、3D都市モデルの整備・活用・オープンデータ化のリーディングプロジェクトである。この PLATEAU のデータは様々な用途に活用できる可能性を持つが、莫大なデータ量のため、使用する PC に高いスペックが求められることや、ダウンロードに時間を要したり、ストレージ容量が必要であるなどの問題があった。 これらの問題は、「PLATEAU SDK for Unity」を使用することで解決することができる。この SDK を使用すると、前述したような問題を軽減しつつ、PLATEAU の豊富なデータを使用して、現実世界のアプリケーションの開発や、都市シミュレーションの開発が可能となる。 そこで本稿では、PLATEAU と Unity の解説をした後、PLATEAU SDK for Unity の導入方法とそのメリットや活用について述べる。

This paper describes an integrated learning system for first-year students to learn basic computer skills, including automated grading modules for typewriting and MS-Excel files and MS-Word files. The system aims to relieve teachers’ workloads to grade many MS-Excel and MS-Word files. It also provides immediate feedback and has a mechanism to prevent students from submitting copied files. In addition, this paper describes the time to grade typewriting, MS-Excel, and MS-Word files. It computes the students’ average normalized gain by using the operational records of the system in our university in 2021. The average normalized gain shows the variation between students’ computer skills decreased. These results, therefore, indicate the effectiveness of the system.