Posts tagged Reinforcement learning
There is an interesting must watch Video of Paul Phillips discussing dopamine based learning and I have embedded the video below. Please do have a look. Original video with some other relevant information can be found here.
Paul discusses the dopamine surge at the Unconditioned Stimulus/reward (US) before the training being replaced by a dopamine surge at Conditioned stimulus(CS) after the learning and how this has led to the reward prediction theory (Schultz) of dopamine function. He also discusses the Cambridge(UK) or Ken Berridge group of objections to this and their discovery that different regions of the brain react differently (ventral and dorsal striatum have different dopamine surges associated with the same stimulus/reward pair in the same animal). He uses electrochemical methods (suing electrodes implanted in the rats brains) to measure tonic dopamine release and there are interesting and informative graphics as well as videso of rats indulging in approach behavior as soon as CS is presented (after conditioning).
He also discusses model-free and model-based reinforcement learning paradigms and discusses how dopamine is only necessary for model-free (simple value association) learning and is not necessarily involved in model-based learning. this he demonstrates beautifully with videos and graphs of selectively bred mice (bred for locomotion), in which high locomotors follow a different (sign-based or model-free) while the low locomotors display goal based or model-based learning. By dopamine manipulations (giving a dopamine antagonist) he is able to show that dopamine is not really necessary for model-based learning.
All in all a very engaging and informative video. A must, must watch that is very highly recommended for anybody who has given a minor thought also to dopamine and what it does in the brain.