Thanks for the link to the paper on a functional explanation of what dreams do! Do you, by any chance, have more recommendations on what to read on this?
I’ve had fuzzy intuitions that dreams might have something to do with using collected episodes to imagine new ones and do some sort of learning on those (eg maybe improving heuristics/making them more coherent with others/the knowledge or something), but I haven’t properly thought about this or even googled what do people already know, and would be pretty curious if there’s a good collection of links
The usual RL approach to dreams is to treat them as either a kind of exploration (not necessarily model-based, possibly just chopping up memories of episodes to stitch together or doing something intermediate like a successor representation) or focus on the hippocampus which appears to encode spatial representations of memories (which makes sense given agents are embodied and typically must move through space to accomplish anything, as opposed to extremely unnatural behaviors like sitting at a computer all day*) & then be doing something akin to experience-replay to finetune itself.
* You ever notice how rarely dreams are about computers or smartphones? When I kept a dream journal, I was bemused at how completely unrepresentative of my life it was, primarily in omitting computers; my dreams had plenty of time for all sorts of other boring stuff, like going to school having forgotten to do my homework the night before, however… It’s a bit like how fiction struggles to incorporate computers at all. “Here is our contemporary protagonist who spends 10 hours a day on her computer and/or tablet and/or smartphone. Let’s, uh, have her lose it in the first 10 minutes so we can get the plot going.”
Weirdly enough I dream tons about my phone and laptop. It’s usually how I know if I fell asleep or not(while napping during the day). Do I remember looking at my phone or not.
Thanks for the link to the paper on a functional explanation of what dreams do! Do you, by any chance, have more recommendations on what to read on this?
I’ve had fuzzy intuitions that dreams might have something to do with using collected episodes to imagine new ones and do some sort of learning on those (eg maybe improving heuristics/making them more coherent with others/the knowledge or something), but I haven’t properly thought about this or even googled what do people already know, and would be pretty curious if there’s a good collection of links
The usual RL approach to dreams is to treat them as either a kind of exploration (not necessarily model-based, possibly just chopping up memories of episodes to stitch together or doing something intermediate like a successor representation) or focus on the hippocampus which appears to encode spatial representations of memories (which makes sense given agents are embodied and typically must move through space to accomplish anything, as opposed to extremely unnatural behaviors like sitting at a computer all day*) & then be doing something akin to experience-replay to finetune itself.
Maybe look through https://www.reddit.com/r/reinforcementlearning/search?q=flair%3APsych&restrict_sr=on&include_over_18=on&sort=relevance&t=all
* You ever notice how rarely dreams are about computers or smartphones? When I kept a dream journal, I was bemused at how completely unrepresentative of my life it was, primarily in omitting computers; my dreams had plenty of time for all sorts of other boring stuff, like going to school having forgotten to do my homework the night before, however… It’s a bit like how fiction struggles to incorporate computers at all. “Here is our contemporary protagonist who spends 10 hours a day on her computer and/or tablet and/or smartphone. Let’s, uh, have her lose it in the first 10 minutes so we can get the plot going.”
Weirdly enough I dream tons about my phone and laptop. It’s usually how I know if I fell asleep or not(while napping during the day). Do I remember looking at my phone or not.