Beyond Reward Suppression: Near-Optimal Offline Attacks on Warm-Start Bandits with Bounded Rewards

Key points

• Existing attacks typically achieve this by suppressing non-target arms.1
• We study this gap through bounded offline attacks on warm-start bandits, where an attacker can inject only valid action-reward pairs into the warm-start history before deployment.1
• We show that target promotion is not merely a heuristic: when the target arm lies near the lower reward boundary, any order-optimal-cost attack against UCB that makes it selected in nearly all online rounds must allocate a nonvanishing fraction of its cost to the target arm.1
• We further extend the attack to Thompson Sampling, $ε$-greedy, and a broader class of bandit algorithms.1

Sources (1)

• [1]Beyond Reward Suppression: Near-Optimal Offline Attacks on Warm-Start Bandits with Bounded Rewards
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 12:57 PM

  Adversarial attacks on bandits aim to mislead a learner toward a target arm while keeping the attack cost small.

  Existing attacks typically achieve this by suppressing non-target arms.

Extractive summary: sentences quoted from the sources.