京都橘大學

Drawing-based social deduction games require artificial intelligence (AI) agents to infer semantic information from incomplete and evolving visual inputs under asymmetric information conditions. In this study, we address the problem of recognizing drawing targets from partial sketch sequences toward constructing an AI player for Drawing Werewolf, a collaborative drawing game derived from the Werewolf (Mafia) genre. Using stroke-based representations from the "Quick, Draw!" dataset, we formulate incremental sketch classification as a sequence modeling task and compare a unidirectional Long Short-Term Memory (UniLSTM) model with a Transformer-based model under realistic online inference constraints. Experiments were conducted on 44 animal classes, evaluating classification accuracy at different drawing stages ranging from one stroke to completed sketches. The results demonstrate that both models improve as additional strokes are observed; however, the Transformer consistently outperforms UniLSTM across all stroke counts. The performance gap is particularly pronounced in early-stage prediction, where sketches are highly incomplete and ambiguous. Class-wise analyses further reveal that the advantage of self-attention depends on visual characteristics and drawing progression. These findings indicate that self-attention mechanisms are well suited for modeling partial sketch sequences and provide valuable insights for designing AI players capable of real-time inference in drawing-based social deduction games.

Background/Objective: Cognitive frailty, the coexistence of physical frailty and cognitive impairment, is a potentially reversible and high-risk state for dementia. This study examined the association between Color Kanji Pick-out Test (CKPT) app performance and cognitive frailty independent of Mini-Mental State Examination (MMSE) scores in community-dwelling older women. Methods: In this cross-sectional study, the participants were 102 community-dwelling older women without dementia and with MMSE scores ≥ 27 (73.6 ± 6.0 years). Reversible cognitive frailty was defined as subjective cognitive decline (≥1 point in the cognitive domain of the Kihon Checklist) plus physical frailty or prefrailty, according to the Japanese Cardiovascular Health Study (J-CHS) criteria. Firth’s penalized logistic regression using three prespecified models, adjusted for age and education, was used to examine the independent associations between CKPT app performance and MMSE scores with reversible cognitive frailty. Results: Fourteen participants (13.7%) met the criteria for cognitive frailty. In separate models, higher CKPT app and MMSE scores were significantly associated with lower odds of cognitive frailty (CKPT: odds ratio [OR] 0.470, p = 0.019; MMSE: OR 0.548, p = 0.020). In a multivariable model including both measures, the CKPT app (OR 0.499, p = 0.031) and MMSE scores (OR 0.553, p = 0.031) remained independently associated with cognitive frailty, and this model had the lowest Akaike information criterion. Conclusions: The CKPT app performance was independently associated with cognitive frailty beyond global cognition. The CKPT app may detect subtle executive and attentional vulnerabilities not captured by the MMSE, supporting practical, objective, early screening and risk stratification of cognitive frailty.

Courage involves pursuing a valued action despite fear, hesitation, or competing motives. However, the pre-action conflict relevant to judging an act as courageous is difficult to observe directly and manipulate systematically when another human serves as the model. Robots provide a means of representing such otherwise covert motives in a controlled and reproducible form. We therefore examined how a robot’s expression of approach-avoidance conflict is perceived and how it relates to observers’ immediate self-evaluations of personal courage. Speech bubbles were projected near the head of a cylindrical robot to present approach and avoidance motives in a scenario in which the robot encountered a person littering in a park. Study 1 compared conflict and no-conflict conditions and sequential and simultaneous presentation methods to determine whether a robot expressing approach-avoidance conflict was perceived as courageous. Study 2 varied whether the robot expressed one motive direction or both approach and avoidance motives and whether it admonished the litterer, and examined whether post-stimulus courage self-evaluations differed according to observers’ preexisting courage tendency. In Study 1, the conflict condition yielded higher courage and conflict ratings than the no-conflict condition. In Study 2, the predicted three-way interaction was not supported. Instead, we observed an interaction between preexisting courage tendency group and conflict expression that was not specified in the primary hypothesis, with self-evaluations in the conflict condition tending to be higher in the low-courage group and lower in the high-courage group. Although neither group’s conflict contrast was significant after correction for multiple comparisons, this contrasting pattern is an exploratory finding suggesting that robot-expressed conflict may be received differently depending on observers’ preexisting courage tendency. The study provides a controlled HRI paradigm for examining how robot-expressed motives are perceived and raises the possibility that user characteristics may be relevant when designing robots that display cues to internal processes.