3 ms·

I thought this was already understood? Are anyone working on an analogous way of integrating reinforcement learning into training generative models based on wh

I thought this was already understood? Are anyone working on an analogous way of integrating reinforcement learning into training generative models based on what they have experienced in user interactions? Seems like having models “dream” e.g. do hypothesis testing and correct the weights based on the results could be a way out of the lack of enough training data.

calf 2y ago
The mainstream info is predominantly that dreams are basically nonsense content (e.g. people dream they can fly) when executive functions are turned off and brain is doing garbage collection and memory consolidation, whereas this title argues the brain is nevertheless learning causal information from dreams which would suggest a more active role than previously thought.
- fluoridation 2y ago
  I'm with the GP, I thought I had heard this somewhere before. Also, I don't think both propositions are in conflict. Dreams may be meaningless and the brain may be using them for something.
- RoyalHenOil 2y ago
  My hunch (based on my personal experiences) is that these occur in different phases of sleep. I find that while I am falling asleep, I often have indescribable nonsense "hallucinations" (I hesitate to call them that because it makes them sound a lot more cohesive than they really are) that feel almost like random noise. These are not just sensory hallucinations, but also logical hallucinations: I have thoughts and beliefs that make so little rational sense that I don't know how to describe them. But later on I will have complex dreams with coherent characters, plots, etc., and have an inner sense of logic (even though it is dream logic) that I can describe after I wake up.
  - 486sx33 2y ago
    With regard to your “hallucinations” comment, you remind me of something the old hippies told me. I remember asking about acid flashbacks and the heavy users would say oh yes we get flashbacks and I’d ask them lots of questions, eventually I established with them that the most likely time to experience as flashback is falling asleep or waking up from an afternoon nap.
    - anal_reactor 2y ago
      On a very rare occasion my brain gets stuck for a few minutes in the state between sleep and consciousness and this kinda does feel like acid.
  - crazydoggers 2y ago
    Hallucinations and lucid dreams during falling asleep are called hypnagogia and definitely are different as they don’t involve REM. I think a lot of people don’t talk about it for fear of sounding crazy. https://en.m.wikipedia.org/wiki/Hypnagogia https://en.m.wikipedia.org/wiki/Hypnagogia And for those interested In hallucinations, Oliver Sacks has an amazing book that I think also destigmatizes them and covers how different they can be from hypnagogic, to migraine, to drug induced, etc. https://en.m.wikipedia.org/wiki/Hallucinations_(book) https://en.m.wikipedia.org/wiki/Hallucinations_(book)
jonnycomputer 2y ago
I think in RL engineering circles offline, off-policy, generative experience replay is. In neuroscience, the idea is being actively explored, but it isn't a consensus position.

calf 2y ago

The mainstream info is predominantly that dreams are basically nonsense content (e.g. people dream they can fly) when executive functions are turned off and brain is doing garbage collection and memory consolidation, whereas this title argues the brain is nevertheless learning causal information from dreams which would suggest a more active role than previously thought.

fluoridation 2y ago
I'm with the GP, I thought I had heard this somewhere before. Also, I don't think both propositions are in conflict. Dreams may be meaningless and the brain may be using them for something.
RoyalHenOil 2y ago
My hunch (based on my personal experiences) is that these occur in different phases of sleep. I find that while I am falling asleep, I often have indescribable nonsense "hallucinations" (I hesitate to call them that because it makes them sound a lot more cohesive than they really are) that feel almost like random noise. These are not just sensory hallucinations, but also logical hallucinations: I have thoughts and beliefs that make so little rational sense that I don't know how to describe them. But later on I will have complex dreams with coherent characters, plots, etc., and have an inner sense of logic (even though it is dream logic) that I can describe after I wake up.
- 486sx33 2y ago
  With regard to your “hallucinations” comment, you remind me of something the old hippies told me. I remember asking about acid flashbacks and the heavy users would say oh yes we get flashbacks and I’d ask them lots of questions, eventually I established with them that the most likely time to experience as flashback is falling asleep or waking up from an afternoon nap.
  - anal_reactor 2y ago
    On a very rare occasion my brain gets stuck for a few minutes in the state between sleep and consciousness and this kinda does feel like acid.
- crazydoggers 2y ago
  Hallucinations and lucid dreams during falling asleep are called hypnagogia and definitely are different as they don’t involve REM. I think a lot of people don’t talk about it for fear of sounding crazy. https://en.m.wikipedia.org/wiki/Hypnagogia https://en.m.wikipedia.org/wiki/Hypnagogia And for those interested In hallucinations, Oliver Sacks has an amazing book that I think also destigmatizes them and covers how different they can be from hypnagogic, to migraine, to drug induced, etc. https://en.m.wikipedia.org/wiki/Hallucinations_(book) https://en.m.wikipedia.org/wiki/Hallucinations_(book)

jonnycomputer 2y ago

I think in RL engineering circles offline, off-policy, generative experience replay is. In neuroscience, the idea is being actively explored, but it isn't a consensus position.