Robust Testing and Estimation under Manipulation Attacks

Robust Testing and Estimation under Manipulation Attacks (via Zoom)

Abstract: Data from different sources form the backbone of many modern distributed learning systems. Two aspects of such systems that make inference tasks challenging are (i) local information constraints including bandwidth limit and privacy concerns; (ii) possible manipulation attacks due to unreliable sources. Recently, there has been a lot of effort in tackling each challenge individually. In this work, we study inference tasks for discrete distributions where both challenges exist. We relate the strength of manipulation attacks to the earth-mover distance using Hamming distance as the metric between messages from the users. Our lower bounds under local information constraints build on the recent lower bound methods in distributed inference. In the communication constrained setting, we develop novel algorithms based on random hashing and an ℓ1/ℓ1 isometry.

This is joint work with Jayadev Acharya (Cornell) and Huanyu Zhang (Cornell).

Bio: Ziteng Sun is a Ph.D. student in the school of Electrical and Computer Engineering at Cornell University, advised by Prof. Jayadev Acharya. His research interest lies broadly in machine learning, algorithmic statistics, and information theory.  He is particularly interested in studying the tradeoffs between different resources in modern data science, including samples, privacy, communication, memory, and computation. Before joining Cornell, he got a Bachelor of Science degree in Electronic Engineering at Tsinghua University.