DISTAL: Domain Shift for Machine Learning in Internet of Things Applications

Accurate adaptation of theoretical network design for IoT in real-world scenarios, crucial during crises like pandemics, requires generic solutions addressing various issues like node placement and traffic prediction. Machine Learning (ML) solutions must flexibly tackle diverse IoT deployment variables. Current ML approaches for IoT are limited; thus, DISTAL proposes a framework addressing domain shift challenges with innovative ML designs. It introduces an out-ofdistribution (OOD) methodology for wireless networking, utilizes fine-grained datasets from largescale IoT testbeds, conducts holistic evaluations, and validates methodologies in real-world IoT deployments. DISTAL's novelty lies in disrupting traditional approaches by focusing on domain shift, exploiting open-access physical resources, and progressing from theory to lab to practice.