Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Detection of Malicious Agents in Social Learning

Valentina Shumovskaia, Mert Kayaalp, Ali H. Sayed

TL;DR

This letter proposes an algorithm that allows discovering the true state of every individual agent based on the sequence of their beliefs and is also able to locate malicious behavior.

Abstract

Non-Bayesian social learning is a framework for distributed hypothesis testing aimed at learning the true state of the environment. Traditionally, the agents are assumed to receive observations conditioned on the same true state, although it is also possible to examine the case of heterogeneous models across the graph. One important special case is when heterogeneity is caused by the presence of malicious agents whose goal is to move the agents toward a wrong hypothesis. In this work, we propose an algorithm that allows discovering the true state of every individual agent based on the sequence of their beliefs. In so doing, the methodology is also able to locate malicious behavior.

Detection of Malicious Agents in Social Learning

TL;DR

This letter proposes an algorithm that allows discovering the true state of every individual agent based on the sequence of their beliefs and is also able to locate malicious behavior.

Abstract

Non-Bayesian social learning is a framework for distributed hypothesis testing aimed at learning the true state of the environment. Traditionally, the agents are assumed to receive observations conditioned on the same true state, although it is also possible to examine the case of heterogeneous models across the graph. One important special case is when heterogeneity is caused by the presence of malicious agents whose goal is to move the agents toward a wrong hypothesis. In this work, we propose an algorithm that allows discovering the true state of every individual agent based on the sequence of their beliefs. In so doing, the methodology is also able to locate malicious behavior.
Paper Structure (4 sections, 1 theorem, 33 equations, 3 figures, 1 algorithm)

This paper contains 4 sections, 1 theorem, 33 equations, 3 figures, 1 algorithm.

Table of Contents

  1. Introduction and Related Work
  2. Social Learning Model
  3. Inverse Modeling
  4. Computer Experiments

Key Result

Theorem 1

The probability of choosing a wrong hypothesis $\theta \notin \Theta_k^\star$ for agent $k\in\mathcal{N}$ is upper bounded by:

Figures (3)

  • Figure 1: Example of images from the MIRO dataset for classes "bus" and "car".
  • Figure 2: Observation map of each agent.
  • Figure 3: Accuracy of the adaptive social learning strategy bordignon2020adaptive and Algorithm \ref{['alg']}. Yellow represents $\theta_0$, and red represents $\theta_1$. For each fold, social learning accuracy is averaged over the past 100 iterations.

Theorems & Definitions (2)

  • Theorem 1: Probability of error
  • proof