Self-Creating Random Walks for Decentralized Learning under Pac-Man Attacks

The CREATE-IF-LATE (CIL) algorithm is proposed to make random walk algorithms resilient against stealthy termination attacks, often called Pac-Man attacks. It works by having nodes self-create new random walks when they detect a possible termination based on local timing information. This decentralized approach ensures the random walk population does not go extinct and that the resulting stochastic gradient descent still converges.

Open the full episode

More from AI Radio