Large language models hold great promise for transforming K-12 education, but there is an urgent need for systematic evaluation of their core educational capabilities. Existing benchmarks often overlook educational goal cognition and overemphasize answer accuracy, thereby failing to capture deeper subject-level knowledge ability and problem-solving ability. To address this gap, we introduce K-12EduBench: a benchmark for evaluating LLMs’ subject-level knowledge ability, subject-specific problem-solving ability, and educational goal cognition ability in K-12 education. K-12EduBench comprises four components: (1) a dataset of 2,640 objective and 619 subjective questions across nine subjects, annotated with answers, problem-solving processes, and cognitive-level labels; (2) nine Item Response Theory (IRT) models for estimating subject-level knowledge ability; (3) evaluation methods and metrics for assessing multi-step problem-solving ability; and (4) prompts and scoring rubrics for measuring alignment with target cognitive levels. Experiments on advanced LLMs show that education-optimized models consistently outperform general-purpose ones across all three abilities, while under-scaled models lag substantially. We observe a strong positive correlation between subject-level knowledge ability and subject-specific problem-solving ability. Despite gains in educational goal cognition ability, current models—even those tailored for education—still fall short of real-world instructional needs.

Group activity recognition aims to understand the overall activity involving multi-person interactions within complex environments. These activities are intricate, encompassing a multitude of interactions such as communication, coordination, and cooperation among individuals. Identifying and analyzing these interactions poses substantial challenges. Researchers in this field often struggle to integrate diverse information levels, as current methods tend to focus separately on individual behavior, interactional dynamics, and collective patterns. This paper addresses this issue by introducing introduces the Multilevel Information Fusion (MIC-Former) Model, a model that integrates these various aspects into a cohesive framework using the Banzhaf power index ratio from game theory. By fusing trajectory and image data, MIC-Former captures the nuanced intricacies of group activities. The model, segmented into three branches, scrutinizes individual movements, member interactions, and overarching group dynamics via transformer networks. A novel fusion technique, grounded in game theory principles, seamlessly integrates these features. Experimental validation on volleyball and collective activity datasets demonstrates MIC-Former’s superior recognition capabilities, underscoring its potential as a robust tool for unraveling the complexities of group behavior and social interactions.

Single-cell RNA sequencing has revealed cellular heterogeneity in complex tissues, notably benefiting research on diseases such as cancer. However, the integration of single-cell data from small samples with extensive clinical features in bulk data remains underexplored. In this study, we introduce PIPET, an algorithmic method for predicting relevant subpopulations in single-cell data based on multivariate phenotypic information from bulk data. PIPET generates feature vectors for each phenotype from differentially expressed genes in bulk data and then identifies relevant cellular subpopulations by assessing the similarity between single-cell data and these vectors. Subsequently, phenotype-related cell states can be analyzed based on these subpopulations. In simulated datasets, PIPET showed robust performance in predicting multiclassification cellular subpopulations. Application of PIPET to lung adenocarcinoma single-cell RNA sequencing data revealed cellular subpopulations with poor survival and associations with TP53 mutations. Similarly, in breast cancer single-cell data, PIPET identified cellular subpopulations associated with the PAM50 clinical subtypes and triple-negative breast cancer subtypes. Overall, PIPET effectively identified relevant cellular subpopulations in single-cell data, guided by phenotypic information from bulk data. This approach comprehensively delineates the molecular characteristics of each cellular subpopulation, offering insights into disease-related subpopulations and guiding personalized treatment strategies.