Large Language Models (LLMs) have enabled dynamic reasoning in automated data analytics, yet recent multi-agent systems remain limited by rigid, single-path workflows that restrict strategic exploration and often lead to suboptimal outcomes.To overcome these limitations, we propose SPIO (Sequential Plan Integration and Optimization), a framework that replaces rigid workflows with adaptive, multi-path planning across four core modules: data preprocessing, feature engineering, model selection, and hyperparameter tuning.In each module, specialized agents generate diverse candidate strategies, which are cascaded and refined by an optimization agent.SPIO offers two operating modes: SPIO-S for selecting a single optimal pipeline, and SPIO-E for ensembling top-k pipelines to maximize robustness.Extensive evaluations on Kaggle and OpenML benchmarks show that SPIO consistently outperforms state-of-the-art baselines, achieving an average performance gain of 5.6%.By explicitly exploring and integrating multiple solution paths, SPIO delivers a more flexible, accurate, and reliable foundation for automated data science.* denotes equal contribution.† denotes corresponding author(s).

Mild Cognitive Impairment (MCI) affects up to 20% of older adults and often progresses to dementia. While brief cognitive screening tools like the Montreal Cognitive Assessment (MoCA) can aid in early detection and monitoring, their reliance on trained clinicians and susceptibility to test anxiety limit accessibility and ecological validity. Game-based cognitive monitoring presents a promising alternative, yet its sensitivity to cognitive changes in individuals with MCI remains underexplored. This study introduces an analytic pipeline for screening cognitive decline using Neuro-World, a serious gaming platform featuring six adaptive subgames that assess cognitive abilities through metrics such as accuracy and response time. Over 12 weeks, ten participants with MCI completed 24 game sessions. Gameplay data were analyzed using correlation-based feature selection and machine learning models to estimate cognitive function and track longitudinal changes. Results showed strong correlations between game-based assessments and clinician-administered MoCA scores$(r=0.71)$, as well as with longitudinal cognitive changes ($r =0.80)$. These findings highlight the potential of game-based cognitive assessments to provide self-administered, ecologically valid screening for cognitive decline in MCI, supporting early detection in aging populations.

Stroke causes motor impairments in the upper limbs, significantly affecting stroke survivors' ability to perform daily activities. Rehabilitation is critical for motor recovery, and frequent assessments are crucial for monitoring improvements in motor impairment throughout the rehabilitation process. Wearable-based motor assessments offer the potential for frequent and objective monitoring of recovery trajectories. Particularly promising is the movement segmentation technique, which decomposes continuous wrist inertial data into lower-level units of upper-limb movements, grounded in theories of motor control and behavior. However, the technique's responsiveness to changes in motor function in stroke survivors has not yet been extensively studied. In this study, we investigate how inertial sensor data obtained during patients' continuous and task-free upper-limb movements can be processed to monitor the recovery trajectory in subacute stroke survivors as they undergo the recovery process. Our results showed that the variability of morphological shapes of the velocity profile of movement segments significantly correlated with changes in clinical scores on the Fugl-Meyer Assessment for the upper extremity (FMA-UE). Moreover, Receiver Operating Characteristic analysis demonstrated that this feature could distinguish participants who showed improvement based on minimal detectable changes, with an Area Under the Curve of 0.86 for FMA-UE. Our comprehensive analysis demonstrates that the movement segmentation approach offers the potential to support objective and frequent assessment of rehabilitation outcomes.

The utilization of advancing transformer-based deep neural network (DNN) models in edge environments holds the promise of improving productivity for intelligent tasks. However, deploying these models on edge devices with limited resources encounters significant performance challenges. Previous solutions have attempted to distribute computation tasks across devices and perform parallel inferences but often fall short of meeting service-level objectives (SLO). This limitation arises from their inability to effectively harness parallelization in transformer-based models and consider the resource diversity of edge devices. In this paper, we propose Hepti, a practical framework designed to facilitate parallel inference of transformer-based DNN models on heterogeneous edge environments. Hepti is armed with: 1) an understanding of transformer model architecture to enable effective parallel inference and 2) dynamic workload optimization to adapt to changing network and device resource capabilities. Our evaluations confirmed that the Hepti autonomously assesses the resource diversity of edge devices and network status. Furthermore, Hepti achieves a maximum performance improvement of 49.1% and 37.1% compared to the local inference approach and state-of-the-art model parallelisms on the BERT-Large model.