p126

_projects/offline_learning.md

@@ -6,7 +6,7 @@ importance: 2
 category: Learning
 ---
 
-A future aim of the lab is to understand how a lens of offline reinforcement learning may be able to shed light on the consequences of human imagination. Indeed, this is motivated by the seeming paradox that we can learn quickly even though we have few real-world interactions with the states, actions and rewards for a given learning problem. This can be explained by understanding how we use prior knowledge in one-shot learning (i.e., planning), how our brains have learned before our lives begin (i.e., inductive biases conferred via evolutionary-timescale learning) or offline learning (i.e., using a model of the world, or our memories, to simulate internally states, actions, and rewards). Here, we tackle the latter possibility by asking the following questions:
+A future aim of the lab is to understand how [offline learning](https://arxiv.org/pdf/2005.01643) is utilized in human cognition to build plans and accelerate learning. Offline learning can be thought of generally as using our memories to simulate actions. It is "offline" in the sense that we do not receive actual states and rewards from real-world interaction, but instead retrieve those from some model of the world. Offline learning addresses the seeming paradox that humans can learn quickly despite having few real-world interactions with the states, actions and rewards for a given learning problem. This problem could be addressed in other ways, such as using prior knowledge in one-shot learning (i.e., planning), and having strong priors due to evolutionary evolutionary-timescale learning. Here we focus on offline learning by asking the following questions:
 
 - How do beliefs form and consolidate via offline learning?
 - What is the emotional experience engendered during or following offline learning?