Thread by @DeepMindAI

Our new paper, “Reinforcement Learning, Fast and Slow”, reviews recent techniques in deep RL that narrow the gap in learning speed between humans and agents, & demonstrate an interplay between fast and slow learning w/ parallels in animal/human cognition:

cell.com/trends/cogniti…

When episodic memory is used in reinforcement learning, an explicit record of past events is maintained for making decisions about the current situation. The action chosen is the one associated with the highest value, based on the outcomes of similar past situations.

Meta-reinforcement learning quickly adapts to new tasks by learning strong inductive biases. This is done via a slower outer learning loop training on a distribution of tasks, leading to an inner loop that rapidly adapts by maintaining a history of past actions and observations.

These techniques were developed in an AI context, but may have rich implications for psychology and neuroscience, highlighting the deep connections between fast RL and slower, incremental learning in both humans and agents.

We hope this paper will be of interest to anybody curious about the intersection between artificial intelligence research and neuroscience.

The paper is available open access - enjoy!

Matt Botvinick @ritterstorm @janexwang @zebkDotCom @BlundellCharles @demishassabis