Event News

Talk on " Online Robust Mean Estimation" by Sihan Liu (UC San Diego)

We are pleased to inform you about the upcoming seminar by Sihan Liu (UC San Diego) titled:"Online Robust Mean Estimation" Everyone interested is cordially invited to attend!


Online Robust Mean Estimation


The area of robust statistics aims to design estimators for distributions that can tolerate up to a constant fraction of adversarial corruption. Recent advancements in this domain have yielded statistically tight and computationally efficient methods.
We revisit this problem from an online perspective. Concretely, we focus on a scenario where $n$ sensors are measuring some common, ongoing phenomenon. At each time step $t = 1, 2, . . . , T$, the i-th sensor reports its readings $x^{(i)}_t$ for that time step. The algorithm must then output its estimate $\mu_t$ for the mean value of this process at time $t$. We assume that most of the sensors observe independent samples from the process, but an $\epsilon$-fraction of them may have been corrupted and thereby behave maliciously.
We provide two main results within this model. Firstly, we present an efficient algorithm whose error bounds have logarithmic dependency on the time horizon $T$. Secondly, we establish the existence of inefficient algorithms that achieve error bounds independent of T under certain input assumptions, notably the independence of data produced at different rounds.


13:30- / Monday, March. 25th, 2024


Hybrid (Onsite and Online)
Room 1810, NII, and online


If you would like to join, please contact by email.
Email : yyoshida[at]nii.ac.jp