Research Fields
My research interests are in the intersection of Information Theory, Machine Learning, and Network Science. I am particularly interested in fundamental limits of freshness-oriented communications.
Timeliness of Information Transfer in Wireless Networks
Information freshness is a new system design criterion motivated by this observation that information usually has the highest value when it is fresh. For example, think about system status updates or samples that are taken from a Markov process: Once a new update packet or sample packet is given, all older packets and the information they carry become insignificant. Information freshness is quantified by the metric of the age of information (AoI) and it captures both how often information is transmitted and how much delay information experiences in a network. To keep the information freshest is an everlasting goal in time-sensitive applications, such as cyber-physical systems, the Internet of Things, smart cities, as well as healthcare systems. Our research focuses on developing novel strategies that can minimize age in wireless networks and we further investigate fundamental tradeoffs between the AoI and other metrics such as estimation error, communication rate, channel throughput, delay, transmission cost, etc. Selective papers: Age of information in random access channels Real-time sampling and estimation on random access channels: age of information and beyond
Real-time Learning in Networks
We investigate timely inference, detection and control for processes that not only evolve over time but also on an underlying network, such as the spread of an infectious disease in a contact network, the spread of a computer virus on the world wide web, or the spread of opinions/product adaption/misinformation in social networks. How to "contain the spread" (or "maximize the influence") as soon as possible is another instance where timeliness is crucial because ''infected'' nodes who are not ''isolated'' may "infect" others. Our goal is to develop sequential and adaptive learning frameworks for deciding which nodes to ''test'' depending on observations. In general, especially in settings where decision-making is decentralized or based on partial or noisy observations, it remains open to proposing novel learning frameworks. We identify interesting tradeoffs between maximizing reward and optimal information extraction and developing algorithmic frameworks toward understanding such tradeoffs. Selective papers: Decentralized Learning Strategies for Estimation Error Minimization with Graph Neural Networks Containing a spread through sequential learning: to exploit or to explore?