Intransitive indifference is a well-documented phenomenon in which the decision maker is forced to choose between alternatives with a subtle difference. In this paper, we establish a general model that extends the semiorder/interval order approach (e.g., Fishburn in J Math Psychol 7: 144–149, 1970a. https://doi.org/10.1016/0022-2496(70)90062-3 ; Luce in Econometrica 24: 178–191, 1956. https://doi.org/10.2307/1905751 ) by adopting the set of lotteries as the domain of choice and a direction-dependent just-noticeable difference function. The model can distinguish two classes of intransitive indifference, that is, imperfect discrimination, which is relevant to previous studies on intransitive indifference, and uncertainty about tastes, which is relevant to incomplete preferences. The main theorem axiomatizes the essentially unique expected utility with direction-dependent sensitivity representation. The key axioms for this characterization are irresolute independence, wherein mixing alternatives with another alternative may change a strict preference to indifference while preserving indifference, and strict preference convexity, which derives the convexity of strict upper and lower contour sets. We also obtain two special cases of our model—one-directional and categorical sensitivity—which highlight the two classes of intransitive indifference, and discuss a possible change in the domain of choice to a vector space.
Although physical activity is recommended for managing cancer-related fatigue (CRF), activity pacing (AP) and energy conservation are also implemented to balance activity and rest. This systematic review assessed the effectiveness of AP-related interventions and behavior change techniques (BCTs) employed in AP-related interventions for CRF management. We searched PubMed, CINAHL, CENTRAL, and Ichu-shi databases for articles published up to February 28, 2026. After assessing the risk of bias, we performed meta-analyses for quantitative synthesis. We also identified and classified BCTs using data from the Behavior Change Technique Taxonomy version 1. Among the 1257 identified studies, 10 (7 randomized controlled trials [RCTs] and 3 non-RCTs) met the inclusion criteria. The effect size was small but not significant (standardized mean differences [SMD] = − 0.36; 95
The damage initiation and evolution behaviors of ferrite-martensite dual phase (DP), transformation-induced plasticity (TRIP)-aided dual-phase (TDP), quenched and tempered (QT), and TRIP-aided martensitic (TM) steels during tensile deformation were investigated. Voids were initiated at the phase boundaries and inside the martensite in the DP and TDP steels, whereas fine voids were observed at the prior austenite, packet, and block boundaries in the QT and TM steels. In the DP and TDP steels, the size of the voids remarkably increased with the plastic strain, even though the number of voids increased slightly. By contrast, the QT and TM steels exhibited a drastic increase in the number of voids, whereas a slight increase in the size of the voids was observed. The voids in the TM steel hardly extended as the plastic strain increased unlike those in the QT steel. The extent of voids in the DP and TDP steels might be attributed to stress and plastic strain partitioning between the different phases during tensile deformation. In addition, the promotion of void initiation and suppression of void growth might be attributed to the fine and uniform martensite matrix in the QT and TM steels. The suppression of void initiation in the TDP steel and void growth in the TM steel might be attributed to the stress and plastic strain relaxations at the void initiation site and the vicinity of voids owing to the effective martensitic transformation of retained austenite.
Large language models (LLMs) have achieved remarkable performance across diverse tasks but remain computationally intensive, limiting their deployment on generalpurpose computing environments. This study investigates model compression techniques, specifically adapter-based fine-tuning and post-training quantization, to enable efficient operation of LLMs under resource constraints. Using the Meta Llama 3.18B Instruct model as a baseline, we evaluate two adapter methods, LoRA and QLoRA, and two quantization techniques, GPTQ and AWQ. Benchmarking on MMLU and JMMLU datasets reveals that QLoRA achieves the highest accuracy among compressed configurations, while GPTQ applied after LoRA maintains performance with reduced model size. Our findings indicate that combining adapter-based fine-tuning with posttraining quantization reduces model size and memory requirements while maintaining competitive accuracy. Although direct measurements of inference speed and power consumption are outside the scope of this study, the observed reduction in model complexity suggests potential applicability to environments with limited computational resources such as general-purpose lowpower devices.
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.
