Cats have all the strengths (& weaknesses) of DRL (esp DQN): generalized, sample-inefficient, perverse genius of trial-and-error, sleep()s all the time while blocking (the door), poor exploration requiring expert demonstrations, reward hackers, hairy to train, black blobs...
-
Show this thread
-
...inability to generalize, short-term credit assignment only, lack of modelling or planning, overly-averse to states which were harmful without realizing which disentangled factor was responsible (cat/hot stove etc)... I've started thinking of my cat as a DQN & it seems to help.
1 reply 2 retweets 23 likesShow this thread -
(Once you see it, you can't go back to modeling your cat as a little human in a fur suit. On the other hand, I suspect for most people it would be more useful to go the other direction and analogize DQNs to cats...)
2 replies 0 retweets 12 likesShow this thread
Conditioning example: When I got Oolong I got him some Purina tuna treats which came in a blue bottle. A year or two later, couldn't find replacement, so bought others. He disdained them all despite being a variety of flavors/brands. Solution: dump'em into the bottle. He's happy.
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.