Recent advances in Time Series Foundation Models (TSFMs) have fundamentally revolutionized general time series analysis across domains like finance, retail, weather, and power. However, how to unlock the hidden capacity of general-purpose TSFMs for wearable activity recognition still remains largely unexplored, given severe sensor annotation scarcity and highly heterogeneous sensor data. To address these challenges, we propose DeepSenseMoE—a novel multi-scale convolution-based Mixture of Experts (MoE) module for parameter-efficient fine-tuning of general-purpose TSFMs to sensor-based activity recognition. DeepSenseMoE integrates three key innovations: (1) Multi-scale convolutional experts with different filter sizes responsible for capturing varying sensor contexts; (2) Shared-expert isolation mechanism compressing common activity knowledge into a single shared expert while reducing redundancy among routed experts; and (3) Hierarchical supervised contrastive alignment guiding experts to further learn discriminative activity features. Extensive experiments on three challenging HAR benchmarks demonstrate DeepSenseMoE's superiority, achieving up to 9.5% accuracy gains over state-of-the-art under few-shot and full-supervised settings, with only

Large Language Models (LLMs) often fail to utilize their latent reasoning capabilities due to a distributional mismatch between ambiguous human inquiries and the structured logic required for machine activation.Existing alignment methods either incur prohibitive O(N ) costs by fine-tuning each model individually or rely on static prompts that fail to resolve query-level structural complexity.In this paper, we propose ReQueR (Reinforcement Query Refinement), a modular framework that treats reasoning elicitation as an inference-time alignment task.We train a specialized Refiner policy via Reinforcement Learning to rewrite raw queries into explicit logical decompositions, treating frozen LLMs as the environment.Rooted in the classical Zone of Proximal Development from educational psychology, we introduce the Adaptive Solver Hierarchy, a curriculum mechanism that stabilizes training by dynamically aligning environmental difficulty with the Refiner's evolving competence.ReQueR yields consistent absolute gains of 1.7%-7.2%across diverse architectures and benchmarks, outperforming strong baselines by 2.1% on average.Crucially, it provides a promising paradigm for one-to-many inference-time reasoning elicitation, enabling a single Refiner trained on a small set of models to effectively unlock reasoning in diverse unseen models.

Inspired by recent success of foundation models in vision and language domains, time series foundation models (TSFMs) have garnered increasing attention in general time series analysis tasks like finance, weather, healthcare, and power. However, given high heterogeneity and severe annotation scarcity in time series sensor data, how to unlock the potential of large-scale general-purpose TSFMs for downstream activity recognition tasks remains yet unexplored? This paper makes the first attempt to address this timely challenge by adapting the self-supervised pre-trained TSFM (i.e., MOMENT) to few-shot activity recognition. We introduce a simple and efficient Sensor-Prompt Tuning (SPT) strategy, which employs multiple convolution-based sensor-friendly filters with a gating mechanism to act as learnable soft prompts, which can dynamically adapt sensor input space to the frozen TSFM backbone, effectively bridging domain gap between pre-training general time series data with wearable sensor stream. Extensive experiments across three public activity recognition benchmarks demonstrate that our SPT achieves up to 15.5% performance gains over existing state-of-the-art baselines under few-shot scenarios, while considerably outperforming other mainstream fine-tuning strategies with smaller than 1% of backbone parameters. Practical cloud-edge inference latencies are measured. This work offers a new prompt-tuning perspective on how to adapt pre-trained TSFMs for wearable activity recognition tasks. Code will be released.

During recent years, dynamic early-exit has provided a promising paradigm to improve the computational efficiency of deep neural networks by constructing multiple classifiers to let easy samples exit at shallow layers while avoiding redundant computations at deep exits, which has been seldom explored in the context of latency-aware human activity recognition (HAR) deployed on wearable devices. Particularly, most existing early-exit strategies have always treated all activity samples equally at each exit during training, which ignore such dynamic early-exit behavior at test-time, causing a potential mismatch between training and test. Intuitively, easy activity samples that often exit earlier at test-time should place more emphasis on the training loss of shallow classifiers, while hard activity samples should contribute more to the training loss of deep classifiers. To bridge this gap, this paper introduces a sample-reweighting approach for efficient activity inference, which employs a weight-predicting network to reweight the training loss of different activity samples at every exit. From a perspective of meta learning, a new optimization objective function is designed to jointly optimize both weight-predicting network and backbone network. We perform extensive experiments on three popular HAR benchmarks including UCI-HAR, WISDM, and UniMiB-SHAR, which demonstrate that while incorporating such test-time early-exit behavior into conventional training pipeline, it can consistently improve the accuracy-efficiency trade-offs under budgeted batch classification and anytime prediction patterns. Moreover, our approach has a natural advantage in handing class-imbalance HAR problem. Detailed ablation studies, visualized illustrations, and real hardware deployment are provided to support our statement.

Self-supervised Human Activity Recognition (HAR) has been gradually gaining a lot of attention in ubiquitous computing community. Its current focus primarily lies in how to overcome the challenge of manually labeling complicated and intricate sensor data from wearable devices, which is often hard to interpret. However, current self-supervised algorithms encounter three main challenges: performance variability caused by data augmentations in contrastive learning paradigm, limitations imposed by traditional self-supervised models, and the computational load deployed on wearable devices by current mainstream transformer encoders. To comprehensively tackle these challenges, this paper proposes a powerful self-supervised approach for HAR from a novel perspective of denoising autoencoder, the first of its kind to explore how to reconstruct masked sensor data built on a commonly employed, well-designed, and computationally efficient fully convolutional network. Extensive experiments demonstrate that our proposed Masked Convolutional AutoEncoder (MaskCAE) outperforms current state-of-the-art algorithms in self-supervised, fully supervised, and semi-supervised situations without relying on any data augmentations, which fills the gap of masked sensor data modeling in HAR area. Visualization analyses show that our MaskCAE could effectively capture temporal semantics in time series sensor data, indicating its great potential in modeling abstracted sensor data. An actual implementation is evaluated on an embedded platform.

Federated Learning (FL) has recently attracted great interest in sensor-based human activity recognition (HAR) tasks. However, in real-world environment, sensor data on devices is non-independently and identically distributed (Non-IID), e.g., activity data recorded by most devices is sparse, and sensor data distribution for each client may be inconsistent. As a result, the traditional FL methods in the heterogeneous environment may incur a drifted global model that causes slow convergence and a heavy communication burden. Although some FL methods are gradually being applied to HAR, they are designed for overly ideal scenarios and do not address such Non-IID problem in the real-world setting. It is still a question whether they can be applied to cross-device FL. To tackle this challenge, we propose ProtoHAR, a prototype-guided FL framework for HAR, which aims to decouple the representation and classifier in the heterogeneous FL setting efficiently. It leverages the global prototype to correct the activity feature representation to make the prototype knowledge flow among clients without leaking privacy while solving a better classifier to avoid excessive drift of the local model in personalized training. Extensive experiments are conducted on four publicly available datasets: USC-HAD, UNIMIB-SHAR, PAMAP2, and HARBOX, which are collected in both controlled environments and real-world scenarios. The results show that compared with the state-of-the-art FL algorithms, ProtoHAR achieves the best performance and faster convergence speed in HAR datasets.