causal learner modeling

Learning multiple subtask skills efficiently for complex robotic manipulation remains challenging. Method: This paper proposes a segment-wise skill learning framework based on latent-space modeling. It formalizes human demonstrations as a latent-variable-driven skill segmentation process, reinterprets mixture density networks (MDNs) for the first time as a library of feedback controllers conditioned on latent states, and constructs a unified probabilistic graphical model integrating skill segmentation and control law inference. Theoretical synthesis of linear feedback control and behavioral cloning enables joint optimization of skill identification and robust control in the latent space. Results: Experiments demonstrate significant improvements in task success rate and robustness to observation noise. The method is validated on real robotic platforms, confirming deployment stability and cross-task generalization capability.

Existing online learning analytics struggle to capture the nonlinear dynamics of student ability, often requiring a prespecified number of clusters and failing to effectively model the relationship between engagement behaviors and ability evolution. This work proposes a Bayesian nonparametric dynamic item response theory framework that employs B-spline basis functions to flexibly characterize the nonlinear influence of participation on ability drift. By incorporating a Mixture-of-Finite-Mixtures prior, the model automatically infers the number of latent learner subgroups, enabling unsupervised clustering and longitudinal tracking of individual ability trajectories. Applied to data from 198 undergraduate students in a statistics course, the model identified four distinct learner types—struggling-declining, low-stable, mainstream-stable, and high-improving—revealing highly stable ability trajectories and no significant predictive effect of participation volume on ability drift.

This work addresses the insufficient coupling between system-level collective behavior and individual dynamics in multi-agent collaborative time-series modeling. We propose a hierarchical recursive switching state model, featuring a two-layer hidden Markov–recurrent coupled architecture that enables context-aware bottom-up and structure-driven top-down latent-state interactions—marking the first explicit characterization of topological influence of group dynamics on individual trajectories. Learning is performed via variational coordinate ascent, ensuring linear scalability in unsupervised training. Empirically, our model matches the predictive accuracy of large neural networks on basketball and military coordination datasets, while employing orders-of-magnitude fewer parameters and exhibiting linear training cost growth with respect to agent count. Furthermore, it successfully uncovers phased collaborative patterns in a synthetic 64-agent task, demonstrating interpretability and scalability in complex multi-agent dynamics.

This study addresses the challenge of accurately tracking students’ dynamic mastery of specific skills when the Q-matrix is unknown. Building upon dynamic cognitive diagnosis models, it compares a joint estimation approach—simultaneously inferring the Q-matrix and learning trajectories—with a two-step strategy that first estimates the Q-matrix and then analyzes skill development. Leveraging reading game data and item text embeddings, the research investigates vocabulary and comprehension growth among second- to third-grade students. The authors propose a bias-corrected two-step method and use simulation studies to delineate the conditions under which each approach performs best: joint modeling proves more reliable when the Q-matrix is uncertain and items vary across grade levels. Empirical results indicate that both methods identify a general trend toward mastering both skills, yet they diverge in estimating the proportion of partial mastery in third grade, underscoring the substantive impact of modeling choices on diagnostic conclusions.

This work addresses the lack of systematic methods for translating expert knowledge into actionable feedback in real-time strategy (RTS) games. It introduces, for the first time, the champion model framework from sports science into RTS training, leveraging 23,305 professional StarCraft II matches to construct a latent performance space via a guided variational autoencoder. The study proposes four counterfactual traversal strategies—linear interpolation, iterative optimal transport, density-regularized gradient ascent, and neural flow matching—to generate multi-step improvement trajectories. Evaluated on out-of-distribution amateur gameplay data, the approach effectively produces interpretable, multi-granular feedback that closely aligns with expert behaviors and leads to victory, thereby enabling algorithmic coaching aimed at human skill enhancement.