We propose novel takes on online and distributed learning with applications to decarbonization, healthcare and communications. This convergent research program hews together engineering, computer science, social science and medicine, and is dictated by pressing needs of Hawaii as well as challenging problems in AI and data science. An unique aspect of our work in these areas is the presence of dynamic data, often incomplete, from distributed sources. We therefore advance

  • online learning in dynamic environments, using our novel framework that tunes stability and adaptability of learned models via regularization; and which extends to risk management, adapting hyperparameters in a principled way and has resolved a number of open problems in online classification and hypothesis testing thus far, and

  • distributed and federated learning, where we propose a kernel based, adaptable learning framework.

Armed with these approaches, we develop a deeper understanding of inference and decision making that combines information from learned models with physical domain knowledge and other constraints, be it economic or social. Specifically, we build on

  • reinforcement learning exploiting imperfect model side information
  • learning in semi-supervised setups with online kernel signal processing approaches,
  • anomaly detection and hyperparameter adaptation using Minimum Description Length (MDL) like approaches, and
  • causality in economics experiment design informed by machine learning best practices.

This fundamental research is used to juxtapose engineering and AI in the context of economic and social considerations in

  • Decarbonization, focusing on the energy and transportation sectors, particularly relevant to Hawaii with its high penetration of renewables and associated challenges. Building on UH Manoa research, we take advantage of collaborations with local stakeholders and utilities for granular and high-frequency electricity usage and transportation data; and the UHM campus, which has come to be a natural living lab to collect data and conduct experiments. We adapt our online learning framework to produce adaptable, faster response machine learning predictors in real systems. We investigate targeting policies to the right groups using a combination of tools we have developed, as well as behavioral explorations in the transportation sector to reduce greenhouse emissions. Hawaii is already a leading state in clean energy integration, and we expect our work to provide insights on how to enhance equitable and inclusive clean energy transitions in the nation and beyond.

  • Healthcare, particularly disease detection from ECGs and epidemics modeling, leveraging perhaps the largest pediatric ECG dataset (collected and curated by one of the PIs). Native Hawaiians and Pacific Islanders are affected by heart diseases at a higher rate, and the unique geography of Hawaii meant that pandemic dynamics, such as the spread of COVID-19 differed from how diseases would spread on the mainland. Here again we combine domain knowledge and incomplete, possibly biased measurements within our broader framework of anomaly detection using generative approaches.

  • Communications, Networks and Security, a central tenet of a sustainable economy in Hawaii, made essential by its geographical isolation. Our work focuses on creating NextG wireless systems with enhanced communication, novel sensing, attack resiliency and energy efficiency. With the evolution of cellular systems to mmWave and sub-THz bands, we focus on sensing solutions using mobile communications as a sensor, for which we develop algorithms that leverage the spatial diversity of a multiple antenna systems and frequency diversity of a wideband antenna systems; innovative wireless physical layer security using reconfigurable devices; and tackling coverage issues in the mmWave band using intelligent reflecting surfaces (large number of low cost reflecting elements with reconfigurable parameters).