This reminds me of something I read on reinforcement learning playing Atari / video games ... often you see the Deep net (I think the game in question was a tank game) just spazzing out on the controller when there’s no enemy, the tank was spinning on the spot while waiting for a door to open for example.
Initially thought just to probably be some numerical instability when the next actions are all equally likely (I.e. there’s nothing for it to do) but what was actually happening is the pseudo RNG was partially fed by controller input, and certain patterns of controller input would make random rewards more likely to spawn.
I always thought it was cool how these exploits get discovered by the deep nets
Initially thought just to probably be some numerical instability when the next actions are all equally likely (I.e. there’s nothing for it to do) but what was actually happening is the pseudo RNG was partially fed by controller input, and certain patterns of controller input would make random rewards more likely to spawn.
I always thought it was cool how these exploits get discovered by the deep nets