Thoughts on software, infrastructure, and building things.
A technical review of a new adversarial learning framework for wearable sensor-based human activity recognition that embeds activity context into the discrimination task, achieving consistent improvements across three HAR benchmarks under leave-one-subject-out evaluation.
New open datasets for contactless neonatal vital signs and industrial activity recognition — what they offer and why they matter.
A unified framework that combines client clustering with adaptive model pruning for federated learning on resource-constrained devices. Evaluated on wearable sensor data with promising results.