Deep Neural Networks (DNNs) have significantly enhanced the baseline performance of Human Activity Recognition (HAR) models by learning patterns directly from raw sensor signals. For wearables, on-device inference is crucial for preserving personally identifiable information and ensuring long battery life, particularly in HAR-based health and wellness applications. However, existing DNN models for HAR are often either too large for wearable devices, designed for simple binary activity classification, or rely on manually crafted features. To address these challenges, we propose a scalable solution comprising a tunable tiny attention condenser-based architecture and a Deep Q-Learning-based tuner. Together, these components enable the generation of compact, wearable-friendly DNN models for complex activity recognition tasks. Experimental results demonstrate that the proposed methodology achieves state-of-the-art accuracy on the HAR-Box and UCI-HAR datasets, with model sizes under 1 MB, making it suitable for resource-constrained devices.

Stress management is paramount to maintaining optimal health and well-being; stress builds up in spikes, causing problems like hypertension and anxiety, necessitating personalized interventions delivered in real-time through wearable technology. This work underscores the pivotal role of unobtrusive stress detection and presents the development of auto-generated compact models tailored for on-device inference through Neural Architecture Search (NAS). These models aim to facilitate efficient stress monitoring directly on low-power devices, representing a promising avenue for advancing personalized healthcare with continuous monitoring and effective digital interventions.

Customized, on-device applications that provide timely interventions about smoking episodes are very helpful for smoking cessation. For this, real-time detection of smoking puffs are necessary through unobtrusive wearable devices. This work demonstrates auto-generated tiny puff detection models for on-device inference on low-power wearable devices.

In this work, we present a lightweight attention network to perform Time Series Classification on Edge devices. We evaluate the merit of our system on a Human Activity Recognition dataset and show the demonstration with the help of a Wearable device (Smartwatch) with IMU sensors.

Continuous monitoring of cardiac health through single-lead wearable Electrocardiogram (ECG), is important for paroxysmal Atrial Fibrillation (AF) detection. Wearable ECG straps, watches, and implantable loop recorders (ILR) are based on this paradigm. These devices are used by medical professionals to view data from multiple patients, perform continuous monitoring and analysis to provide immediate care to patients. These monitoring devices display simple health screening alerts to the subjects and generate distress signals for people working outdoors or in isolated environments with intermittent Internet connectivity. Hence, low-memory, low-power, low-latency on-device inference becomes very important. This work aims at realizing such solutions by providing a framework to generate tiny (less than 256 KB) Deep Neural Networks customized for typical microcontrollers (MCU) used in those devices.

Wearable health monitoring has become a very familiar term in today'sworld. One of the most popular means ofwearable sensing is photoplethysmogram (PPG). Due to its unobtrusive and ubiquitous nature, it is gaining popularity among people everywhere. Due to the ease of use, the utility of such technology is increasing day by day. However, in theworld of researchers, the accurate estimation of heart rate (HR) in presence of motion artefacts remains an unsolved problem due to the susceptibility of PPG signals to corruption by motion artefacts. The way in which a person fastens the device on the wrist plays an important role in the acquisition of signal from the device. While there are various research works going on in this field, there is always a trade-off between accuracy and complexity of algorithm and hardware resources. Also, in such scenarios where the sensor gets misplaced due to movements, there might be no PPG signal component available in the acquired signal data. In such cases the sophisticated denoising algorithms make no sense.