Age-of-Information — Robust Analysis & Pull-Model Learning
New analytical tools (robust queueing) and learning-based methods for characterizing and optimizing information freshness.
Beyond designing update and scheduling policies, my research develops new analytical and learning-based tools for Age-of-Information (AoI). Classical AoI analysis relies on probabilistic assumptions (e.g., memoryless or i.i.d. arrivals) that break down for the heavy-tailed and correlated traffic seen in practice; and in many modern applications the freshness-optimal action must be learned online rather than computed from a known model.
Robust-queueing analysis of AoI. I introduced a robust-optimization approach to bound the Peak AoI (PAoI): instead of assuming a specific distribution, I model the uncertainty in the arrival and service processes with uncertainty sets, yielding worst-case PAoI approximations for very general (including heavy-tailed and correlated) processes. Unlike classical bounds (e.g., Kingman-type bounds under i.i.d. assumptions) that are loose under light load, the resulting approximation is accurate across both light and high load—the two regimes most critical for AoI.
Freshness under the Pull model with learning. I also studied a Pull model in which a user proactively sends replicated requests to multiple servers to fetch fresh information. Replication exposes a novel tradeoff between the differing AoI values and response times across servers; I derived the optimal number of responses to wait for and showed that—perhaps counterintuitively—waiting for more than one response often reduces AoI. When the system is unknown, I reformulated freshness (utility) maximization as a stochastic multi-armed bandit with side observations and designed learning algorithms with improved performance guarantees.
Papers (available on the publications page):
[1] Liu, Zhongdong, et al. “A Worst-Case Approximate Analysis of Peak Age-of-Information Via Robust Queueing Approach.” IEEE INFOCOM 2021.
[2] Li, Fengjiao, Yu Sang, Zhongdong Liu, et al. “Waiting But Not Aging: Optimizing Information Freshness Under the Pull Model.” IEEE/ACM Transactions on Networking 29.1 (2021): 465–478.