Faculty Profiles - FUJII Keisuke

Papers - FUJII Keisuke

Division display 1 - 20 of about 130 ／ All the affair displays >>

Estimating Counterfactual Treatment Outcomes Over Time in Complex Multiagent Scenarios. Reviewed

Keisuke Fujii, Koh Takeuchi, Atsushi Kuribayashi, Naoya Takeishi, Yoshinobu Kawahara, Kazuya Takeda

IEEE Trans. Neural Networks Learn. Syst. Vol. 36 ( 2 ) page： 2103 - 2117 2025.2

特集「スポーツ×情報学×インタラクション」にあたって Open Access

市川淳, 藤井慶輔, 山田雅敏, 一ノ瀬元喜

人工知能 Vol. 40 ( 1 ) page： 1002 - 1003 2025.1

Evaluating Timing of Movement Starts Using Counterfactual Virtuality in Ultimate Reviewed Open Access

IWASHITA Shunsuke, FUJII Keisuke

Proceedings of the Annual Conference of JSAI Vol. JSAI2025 ( 0 ) page： 2M4OS37a05 - 2M4OS37a05 2025

Language：Japanese Publishing type：Research paper (scientific journal) Publisher：The Japanese Society for Artificial Intelligence

Ultimate is a sport played by two teams of seven players, where points are scored by passing a disc and catching it in the opposing team’s end zone. A unique rule of this sport is that the player holding the disc cannot move, making field dynamics primarily driven by other players’ movements. In particular, the timing of a receiver’s movement is a critical tactical factor. This study aims to propose a quantitative evaluation metric for the timing of players’ movements. First, game footage was recorded using a drone camera, and preprocessing was conducted to obtain players’ positional data. Next, players’ movements were detected using a rule-based approach, and multiple counterfactual scenarios were generated by shifting the timing of movements. These scenarios were analyzed using a space evaluation metric based on soccer’s pitch control, adapted to reflect the unique rules of Ultimate. By comparing the spatial evaluation values across scenarios, the difference between actual play and the most favorable counterfactual scenario was used to quantitatively assess the impact of movement timing. A model was then developed to propose improved movement timing for better tactical outcomes.

DOI： 10.11517/pjsai.jsai2025.0_2m4os37a05

Open Access

CiNii Research

Velocity Completion Task and Method for Event-based Player Positional Data in Group Sports Reviewed Open Access

UMEMOTO Rikuhei, FUJII Keisuke

Proceedings of the Annual Conference of JSAI Vol. JSAI2025 ( 0 ) page： 2E1GS1002 - 2E1GS1002 2025

Language：Japanese Publishing type：Research paper (scientific journal) Publisher：The Japanese Society for Artificial Intelligence

Recently, new team sports data, including event-based player positional data, has been publicly released in soccer, raising expectations for advancements in analytical methods for scouting and recruitment. However, this data does not include full-time player positional information, making it challenging to calculate player velocities for each event and to apply space evaluation methods. This study proposes a method to complete player velocity using only event-based positional data. Furthermore, we evaluate the applicability of existing space evaluation methods based on the completed velocity. Experiments using soccer data show that neural network methods considering graph structures and temporal information outperform rule-based methods, achieving more minor velocity completion errors and producing space evaluation results closer to complete tracking data.

DOI： 10.11517/pjsai.jsai2025.0_2e1gs1002

Open Access

CiNii Research

Detection of front-door and back-door pitches in baseball and the characteristics that make them effective Reviewed Open Access

MIURA Takimi, FUJII Keisuke

Proceedings of the Annual Conference of JSAI Vol. JSAI2025 ( 0 ) page： 1Win488 - 1Win488 2025

Language：Japanese Publishing type：Research paper (scientific journal) Publisher：The Japanese Society for Artificial Intelligence

Front-door and back-door pitches (hereinafter referred to as “door-type pitches”) refer to laterally breaking balls that move from outside to inside the strike zone. Door-type pitches often induce called strike or weak contact, but they carry the risk of hard hits. However, there are no clear criteria for detecting door-type pitches and their effectiveness has not been verified. This study aims to clarify the requirements for whether door-type pitches are effective in NPB. First, we used data from MLB to construct a machine-learning model that estimates the amount of pitch movement, allowing us to detect door-type pitches in NPB. Next, we tested the effectiveness of door-type pitches and analyzed the relationship between the pitchers’ and pitches’ characteristics and the test results. The results suggest that some door-type pitches may be effective for pitches inducing weak contact, some front-door pitches may be ineffective for slow pitches, and some door-type pitches may be ineffective for pitches that break in the same direction as the pitcher’s throwing arm. These findings can help players and coaches refine and evaluate decision-making of their pitching strategies.

DOI： 10.11517/pjsai.jsai2025.0_1win488

Open Access

CiNii Research

Estimation of temporal changes in basketball players' contribution and compatibility based on a state-space model Reviewed Open Access

YAMADA Kazuhiro, FUJII Keisuke

Proceedings of the Annual Conference of JSAI Vol. JSAI2025 ( 0 ) page： 2E1GS1003 - 2E1GS1003 2025

Language：Japanese Publishing type：Research paper (scientific journal) Publisher：The Japanese Society for Artificial Intelligence

In professional sports, the correct evaluation of player contributions and compatibility is an important issue. Metrics that quantitatively evaluate them can be useful information for team decision-makers. Existing metrics often overlook temporal changes in player contributions and compatibility, which may arise from player growth and other dynamic factors. Therefore, the purpose of this study is to propose quantitative evaluation metrics that take into account temporal changes in player contribution and compatibility. First, we created time-by-time design matrices using play-by-play data from the 2022-23 and 2023-24 seasons of the B.LEAGUE. Next, the parameters of the state-space model were Bayesian estimated from the design matrices, and the evaluation values of the proposed metrics Gaussian Random Walk Plus-Minus (GRWPM) for players, duos, lineups, and teams were calculated based on the estimated parameters. As a result, the proposed metric showed a strong positive correlation with the existing metric, suggesting its validity to a certain degree. We also discussed its properties and confirmed that it provides information that cannot be captured by the existing metrics alone. The proposed metrics are expected to be useful for scouting players, determining annual salaries and contract years, and considering roster and lineup composition.

DOI： 10.11517/pjsai.jsai2025.0_2e1gs1003

Open Access

CiNii Research

Evaluation of Player Behavior in Badminton Doubles Using Deep Reinforcement Learning Reviewed Open Access

NAKAHARA Eito, FUJII Keisuke, KAWASHIMA Hiroaki

Proceedings of the Annual Conference of JSAI Vol. JSAI2025 ( 0 ) page： 2E1GS1004 - 2E1GS1004 2025

Language：Japanese Publishing type：Research paper (scientific journal) Publisher：The Japanese Society for Artificial Intelligence

In badminton doubles, there are no specific roles or positions such as attackers or defenders, which exist in many team sports. Therefore, it is important for each player to understand the game situation and take appropriate actions. On the other hand, previous studies have only proposed methods for evaluating players' overall performance and strokes. In this study, we aim to evaluate whether players are able to take appropriate actions in terms of strokes and formations, based on video footage of badminton doubles matches. Specifically, we propose a method to evaluate players' actions using the action value function of deep reinforcement learning, taking as input the players' poses and shuttle positions detected from past match videos. For the evaluation of actions, we use counterfactual data in which real actions are changed into fictitious actions, and compare the results with those obtained when players performed different actions to evaluate whether players performed actions appropriate to the game situation. The experimental results confirmed that the proposed method can capture and evaluate the tendency of the ideal behavior in a game situation, and demonstrated its effectiveness.

DOI： 10.11517/pjsai.jsai2025.0_2e1gs1004

Open Access

CiNii Research

Tactical Decision Analysis in Soccer Using Reinforcement Learning with Interpretable Low-Dimensional States Reviewed Open Access

IDE Kenjiro, SOMEYA Taiga, KAWAGUCHI Kohei, FUJII Keisuke

Proceedings of the Annual Conference of JSAI Vol. JSAI2025 ( 0 ) page： 2M4OS37a04 - 2M4OS37a04 2025

Language：Japanese Publishing type：Research paper (scientific journal) Publisher：The Japanese Society for Artificial Intelligence

In sports like soccer, modeling and analyzing matches is challenging due to the continuous interactions among multiple players. Rule-based models struggle to capture all players and scenarios, while reinforcement learning models often lack interpretability in evaluation values. This study aims to explore whether a low-dimensional, rule-based state variable reinforcement learning model using spatiotemporal data can effectively analyze soccer tactics. We defined state variables and actions based on interviews with a head coach from the Kanto Soccer Division 1 League. All players and scenarios were modeled, reinforcement learning was conducted, and state-action values (Q-values) were calculated. Results revealed a trade-off between action prediction accuracy and TD error. Although no correlation was found with traditional metrics such as xG or goals, the model effectively evaluated high-risk tactics like counterattacks and defensive breakthroughs compared to the baseline. Furthermore, visualizing Q-values and state variables enabled the interpretation of Q-values and evaluation of individual players. This approach shows promise for enhancing coaching strategies and player assessment.

DOI： 10.11517/pjsai.jsai2025.0_2m4os37a04

Open Access

CiNii Research

Attempting to incorporate elements of open skills into closed-skill learning : A case study using a golf putting task Reviewed

HASEGAWA Yumiko, YAMAMOTO Kota, OKADA Ayako, FUJII Keisuke

Vol. 114 page： 101 - 119 2024.6

Language：English Publishing type：Research paper (scientific journal) Publisher：Faculty of Humanities and Social Sciences, Iwate University

In motor learning, how a target skill is learned is important, and performance results vary depending on practice-related factors. Motor skills, which are classified as closed skills, are typically practiced through self-paced repetition. However, learning a target-following task performed alone may be facilitated by intervention from another person. Therefore, we considered an environment with disturbances such as open skills, even for a closed-skill learning activity such as a golf putting task, and expected that learning would be promoted in an environment with high uncertainty. In this study, we developed an idea based on several previous studies, and examined the effects of the practice of "catching and hitting a ball launched from ball launchers." The participants were four golf novices, two of whom practiced a conventional putting style as the control group. The participants practiced 10 times over a month, during which they took a pretest, midterm test, and posttest. Subsequently, the participants were challenged with two tasks. The participants practiced with approximately 1,000 balls during the study. To evaluate their performance, we used a motion capture device to measure the orientation of their body, kinematics of the putter head, and final ball positions. To measure participants’ sight lines, we proposed a method that utilizes a local coordinate system to efficiently represent and estimate the movement of points. From the results of the sight line analysis using this method, we understood the problems of novices’ alignment (aligning the putter head and body for the target). In addition, we were able to determine how long it takes for golf novices to acquire their approximate movement patterns and the number of days that the absolute error of the final ball position can be kept within 0.2 m ~ 0.4 m. We discuss the impact and future possibilities of incorporating elements of open skills to improve golf-putting skills.

DOI： 10.15113/0002000356

CiNii Research

Decentralized policy learning with partial observation and mechanical constraints for multiperson modeling Reviewed

Keisuke Fujii, Naoya Takeishi, Yoshinobu Kawahara, Kazuya Takeda

Neural Networks Vol. 171 page： 40 - 52 2024.3

Adaptive action supervision in reinforcement learning from real-world multi-agent demonstrations Reviewed

Keisuke Fujii, Kazushi Tsutsui, Atom Sco, Hiroshi Nakahara, Naoya Takeishi, Yoshinobu Kawahara

International Conference on Agents and Artificial Intelligence (ICAART 2024) Vol. 2 page： 27 - 39 2024.2

Estimating counterfactual treatment outcomes over time in complex multi-vehicle simulation Reviewed

Keisuke Fujii, Koh Takeuchi, Atsushi Kuribayashi, Naoya Takeishi, Yoshinobu Kawahara, Kazuya Takeda

30th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems (ACM SIGSPATIAL 2022) 2022.11

Evaluation of soccer players to create scoring opportunities for teammates based on their trajectory prediction Reviewed Open Access

TERANISHI Masakiyo, TSUTSUI Kazushi, TAKEDA Kazuya, FUJII Keisuke

Proceedings of the Annual Conference of JSAI Vol. JSAI2022 ( 0 ) page： 3G4OS15b05 - 3G4OS15b05 2022

Language：Japanese Publishing type：Research paper (scientific journal) Publisher：The Japanese Society for Artificial Intelligence

Soccer is a game in which many players and the ball interact in complex ways. Regarding the quantitative evaluation of soccer attackers, there have been many and few studies on the player with and without the ball, respectively. However, it is still difficult to evaluate an attacking player without the ball and intention to receive it, and to reveal how movement contributes to the creation of scoring opportunities compared to typical (or predicted) movements. In this paper, we evaluate players who create off-ball scoring opportunities by comparing the reference movements generated by trajectory prediction with actual movements. In the proposed method, first, the trajectory is predicted using a graph variational recurrent neural network that can accurately model the relationship between players and predict the long-term trajectory. Next, based on the difference in the existing off-ball evaluation index between the actual data and the predicted trajectory, we evaluate how the actual movement contributes to scoring opportunity compared to the predicted movement as a reference. In the verification, we show that the evaluation of the proposed method is intuitive, using the relationship with the scores with all 18 teams in the Japanese professional soccer league and the example of one game.

DOI： 10.11517/pjsai.jsai2022.0_3g4os15b05

Open Access

CiNii Research

A basic study on the mechanism of group behavior of wild bats using movement pattern measurement and granger causality during nesting Reviewed Open Access

USHIO Kazusa, FUJIOKA Emyo, FUJI Keisuke, HABE Hitoshi, KAWASHIMA Hiroaki, HIRYU Shizuko

Proceedings of the Annual Conference of JSAI Vol. JSAI2022 ( 0 ) page： 3G4OS15b02 - 3G4OS15b02 2022

Language：Japanese Publishing type：Research paper (scientific journal) Publisher：The Japanese Society for Artificial Intelligence

Bats recognize their surrounding environment by processing the echoes of ultrasonic waves emitted by themselves. Many species of bats live in groups, and many individuals emerge together from roosts. In this study, we used high-sensitivity video cameras to measure the flight trajectories of bats emerging from the cave in three dimensions, and investigated their flight trails. As a result, we found there were three behavioral patterns during emerging: exiting the cave, returning to the cave, and some other action. In addition, we applied the Granger causality method (Fujii et al., NeurlPS'21) to analyze the swarm behavior mechanism of emerging bats. The results showed that forward individuals flew in such a way that they were "repulsed" from or "approached" the other individuals. This suggests that bats, which use sound to understand their environment, are also influenced by backward individuals, which cannot be captured visually, suggesting that bats have a unique swarming mechanism that differs from model animals for group behavior, mainly visual animals.

DOI： 10.11517/pjsai.jsai2022.0_3g4os15b02

Open Access

CiNii Research

Learning interaction rules from multi-animal trajectories via augmented behavioral model Reviewed

Keisuke Fujii, Naoya Takeishi, Kazushi Tsutsui, Emyo Fujioka, Nozomi Nishiumi, Ryoya Tanaka, Mika Fukushiro, Kaoru Ide, Hiroyoshi Kohno, Ken Yoda, Susumu Takahashi, Shizuko Hiryu, Yoshinobu Kawahara

Advances in Neural Information Processing Systems (NeurIPS'21) Vol. 34 2021.12

Data-Driven Analysis for Understanding Team Sports Behaviors Open Access

Keisuke Fujii

Journal of Robotics and Mechatronics Vol. 33 ( 3 ) page： 505 - 514 2021.6

Authorship：Lead author,　Last author,　Corresponding author Language：English Publishing type：Research paper (scientific journal) Publisher：Fuji Technology Press Ltd.

Understanding the principles of real-world biological multi-agent behaviors is a current challenge in various scientific and engineering fields. The rules regarding the real-world biological multi-agent behaviors such as those in team sports are often largely unknown due to their inherently higher-order interactions, cognition, and body dynamics. Estimation of the rules from data, i.e., via data-driven approaches such as machine learning, provides an effective way to analyze such behaviors. Although most data-driven models have non-linear structures and high predictive performances, it is sometimes hard to interpret them. This survey focuses on data-driven analysis for quantitative understanding of behaviors in invasion team sports such as basketball and football, and introduces two main approaches for understanding such multi-agent behaviors: (1) extracting easily interpretable features or rules from data and (2) generating and controlling behaviors in visually-understandable ways. The first approach involves the visualization of learned representations and the extraction of mathematical structures behind the behaviors. The second approach can be used to test hypotheses by simulating and controlling future and counterfactual behaviors. Lastly, the potential practical applications of extracted rules, features, and generated behaviors are discussed. These approaches can contribute to a better understanding of multi-agent behaviors in the real world.

DOI： 10.20965/jrm.2021.p0505

Open Access

CiNii Research

Physically-interpretable classification of biological network dynamics for complex collective motions. Reviewed International journal Open Access

Keisuke Fujii, Naoya Takeishi, Motokazu Hojo, Yuki Inaba, Yoshinobu Kawahara

Scientific Reports Vol. 10 ( 1 ) page： 3005 - 3005 2020.2

Language：English Publishing type：Research paper (scientific journal)

Understanding biological network dynamics is a fundamental issue in various scientific and engineering fields. Network theory is capable of revealing the relationship between elements and their propagation; however, for complex collective motions, the network properties often transiently and complexly change. A fundamental question addressed here pertains to the classification of collective motion network based on physically-interpretable dynamical properties. Here we apply a data-driven spectral analysis called graph dynamic mode decomposition, which obtains the dynamical properties for collective motion classification. Using a ballgame as an example, we classified the strategic collective motions in different global behaviours and discovered that, in addition to the physical properties, the contextual node information was critical for classification. Furthermore, we discovered the label-specific stronger spectra in the relationship among the nearest agents, providing physical and semantic interpretations. Our approach contributes to the understanding of principles of biological complex network dynamics from the perspective of nonlinear dynamical systems.

DOI： 10.1038/s41598-020-58064-w

Data-driven spectral analysis for coordinative structures in periodic human locomotion. Reviewed International journal Open Access

Keisuke Fujii, Naoya Takeishi, Benio Kibushi, Motoki Kouzaki, Yoshinobu Kawahara

Scientific reports Vol. 9 ( 1 ) page： 16755 - 16755 2019.11

Language：English Publishing type：Research paper (scientific journal)

Living organisms dynamically and flexibly operate a great number of components. As one of such redundant control mechanisms, low-dimensional coordinative structures among multiple components have been investigated. However, structures extracted from the conventional statistical dimensionality reduction methods do not reflect dynamical properties in principle. Here we regard coordinative structures in biological periodic systems with unknown and redundant dynamics as a nonlinear limit-cycle oscillation, and apply a data-driven operator-theoretic spectral analysis, which obtains dynamical properties of coordinative structures such as frequency and phase from the estimated eigenvalues and eigenfunctions of a composition operator. Using segmental angle series during human walking as an example, we first extracted the coordinative structures based on dynamics; e.g. the speed-independent coordinative structures in the harmonics of gait frequency. Second, we discovered the speed-dependent time-evolving behaviours of the phase by estimating the eigenfunctions via our approach on the conventional low-dimensional structures. We also verified our approach using the double pendulum and walking model simulation data. Our results of locomotion analysis suggest that our approach can be useful to analyse biological periodic phenomena from the perspective of nonlinear dynamical systems.

DOI： 10.1038/s41598-019-53187-1

Dynamic mode decomposition in vector-valued reproducing kernel Hilbert spaces for extracting dynamical structure among observables. Reviewed International journal Open Access

Keisuke Fujii, Yoshinobu Kawahara

Neural networks : the official journal of the International Neural Network Society Vol. 117 page： 94 - 103 2019.9

Language：English Publishing type：Research paper (scientific journal)

Understanding nonlinear dynamical systems (NLDSs) is challenging in a variety of engineering and scientific fields. Dynamic mode decomposition (DMD), which is a numerical algorithm for the spectral analysis of Koopman operators, has been attracting attention as a way of obtaining global modal descriptions of NLDSs without requiring explicit prior knowledge. However, since existing DMD algorithms are in principle formulated based on the concatenation of scalar observables, it is not directly applicable to data with dependent structures among observables, which take, for example, the form of a sequence of graphs. In this paper, we formulate Koopman spectral analysis for NLDSs with structures among observables and propose an estimation algorithm for this problem. This method can extract and visualize the underlying low-dimensional global dynamics of NLDSs with structures among observables from data, which can be useful in understanding the underlying dynamics of such NLDSs. To this end, we first formulate the problem of estimating spectra of the Koopman operator defined in vector-valued reproducing kernel Hilbert spaces, and then develop an estimation procedure for this problem by reformulating tensor-based DMD. As a special case of our method, we propose the method named as Graph DMD, which is a numerical algorithm for Koopman spectral analysis of graph dynamical systems, using a sequence of adjacency matrices. We investigate the empirical performance of our method by using synthetic and real-world data.

DOI： 10.1016/j.neunet.2019.04.020

Applying team strategies for dynamic coordination: A comparative study of expertise using 3-on-3 basketball. Reviewed International journal Open Access

Jun Ichikawa, Masatoshi Yamada, Yutaka Iwaihara, Genki Ichinose, Keisuke Fujii

PloS one Vol. 21 ( 2 ) page： e0343077 2026.2

Authorship：Last author Language：English Publishing type：Research paper (scientific journal)

Humans working as a team can achieve higher performance. Studies in sports science, network science, and machine learning have extracted dynamic physical interaction structures of such coordination in team sports. However, the information processing, such as the application of team strategies, has not been fully discussed. The purpose of this cognitive science study was to investigate the application of team strategies for dynamic coordination across different expertise levels using 3-on-3 basketball. In a field experiment, female players, who were selected as prefecture representatives in Japan (Average experience 19.33 years, [Formula: see text] = 4.09), repeatedly engaged in mini-games. Their previous and current affiliated teams competed in national tournaments. We analyzed the difficulty in anticipating offensive movements for the opponent defensive team, quantified as entropy, using tracking position data. This was compared with female players recorded in a previous study. They affiliated with a university team ranked in the third division of the regional league (Average experience 11.33 years, [Formula: see text] = 3.42). There was no such achievement as those in the high expertise condition. Using a linear mixed model with the significance level ([Formula: see text]), the results showed that the entropy for the key player in the high expertise condition was significantly higher than that in the low expertise after the tips condition, and similar to that in the low expertise before the tips condition. The tips were concise coaching advice regarding coordination related to the crucial role of intervention decision and adjustment. This was also lower than that simulated in the random walk condition, which served as a minimal baseline for scientifically explaining the observed complexity. Our first step study suggests that the movement dynamics at the expert level may be relatively complex, making it difficult for the defensive team to anticipate, and related to the application of team strategies.

DOI： 10.1371/journal.pone.0343077

Open Access

PubMed

researchmap

　 PREV - NEXT 　

<FUJII Keisuke>

Papers - FUJII Keisuke