Hippocampus

YT: How Your Brain Chooses What to Remember

Hippocampus handles episodic memory during waking hours, and in sleep repeats strong memory patterns to neocortex which then stores them for long time.

• Hippocampus is part of brain taking part in episodic memory (memory of sequence of events).
• Hippocampus has two modes (one when awake and when asleep).
• Wakefullness Mode: When awake it tracks where and when things happens
• Sleep Mode: When asleep it replays main events from the day. And stores them in neocortex.
  • Consider a part of hippocampus
    • A surge of "sharp wave" activation arrives there from higher part of hippocampus
    • It tries to wake up (excite) the neurons in that area
    • But there are inhibitor neurons that don't allow all neurons to fire at once
    • So, there is a competition. And the neurons that formed strong bond during the day fire.
  • This surge happens multiple times and in short intervals.
  • And the memory patterns that were strong get replayed and transferred to neocortex
• During sleep, neocortex enters a state where it is receptive to signals from hippocampus
• And when same memory patterns are repeatedly sent by the hippocampus to neocortex, it stores them in its permanent memory

Findings from a paper Replay Comes of Age [pdf] by David J. Foster (2017)

Replay was originally believed to occur mainly during sleep but now we know it happens during wakefullness too. These awake reply have role in planning, decision making and working memory. Thus the perspective has changed on replay. It is less like dreaming and more like active internal processing similar to thought.

Reverse replay: Hypothetical Trajectories & Planning:

Hippocampal place cells participate in sequenced activity patterns known as replay after experience. These awake replay can occur quickly after individual experiences. And they are often in the reverse direction compared to the original experience, and they can depict hypothetical trajectories that were never experienced before. Thus replay represents a model of traversable space and relationships between places, rather than just a record of specific experiences.

Planning in forward direction by simulating all trajectories is inefficient, and thus reverse replay seems to have use in goal directed planning.

Like learning value function:

Reverse replay seems to be specifically modulated by changing reward. Increasing the reward (from dopamine signals) increased the rate of reverse replay and vice versa. This leads to the hypothesis that reverse replay support value learning i.e. associating places with their reward value.

Role of Forward replay is unclear:

Forward replay might be involved in retrieving values or planning future actions. In experiments, the forward replay predicts the future goal-directed trajectories before the animal physically takes those actions. One idea is forward replay might represent planned sequences of actions (called "options" in a reinforcement learning framework).