neural computation underlying memory processing

three major areas in HP

CA1

CA3

DG

Hippocampal circuitry

6 deep layers lumped together, no layer 4. entorhinal cortex is old cortex.

major projection found in HP: tri-synaptic pathway

layer II project to DG granule cell and projection goes to CA3 and to the next dendrite at radiatum. CA3 cell axons project to CA1 dendrite on proximal side and CA1 will send backt odeep layers of entorhinal cortex.

memory formation

sensory driven

formation of experience-specific patterns in CA1/CA3

stored in CA3 recurrent synapses (synaptic plasticity)

theta mode (6-10 Hz oscillations in LFPs)

layer II and III will send information to DG and CA3 activation. layer III send projections to CA1. Major process happening in CA3. CA3 neurons fire in a particular pattern and propogate pattern to CA1 and CA1 will have a specific firing pattern. CA3 is the engine of memory processing in HP.

memory consolidation

CA3 (internally) driven (triggered by noise or cortical input)

compressed reactivation of experience-specific patterns in CA1/CA3

compressed reactivation of experience-specific patterns in the neocortex

ripple mode (100-200 Hz in LFPs during sleep, resting)

input to CA1 to layer 5 and 6. layer III input sent to

CA3 reactivation will send input to CA1 reactivation. Ripple oscillation in CA1 will need CA3 input.

memory retrieval

CA3 (internally) driven, triggered by cortical input (cues)

real-time (theta mdoe) or compressed (ripple mode) reactivation of experience-specific patterns in CA3 and CA1

real-time or compressed reactivation of experience-specific patterns in the neocortex

attractor dynamics formation or retrieval: pattern separation (two patterns and circuit separates them apart) or completion (walk into room even though something changed in room, you still know it is the same room)

DG: pattern separation (large number of neurons; low activity; adult neurogenesis)

CA3 is key in this process.

retrieval: before forming a new memory CA3 already has things there, but if you walk into a new room input will come in from cortex and then in neuronal space, every neuron is indepednent variable. Take all the neurons as a high dimensional neural space, every dot is a neuron firing at a specific rate. When you walk into a new room the input will activate neurons in CA3 and then the input will involve with time of synaptic pathway buildup. in the end it converges to a point. AKA an attractor. once you form a memory of a new room next time oyu step in, you will get an attractor due to connections and buildup.

formation: if you step into a new room, input to CA3 decides if it will be a new memory or not. this dynamic process will determine if this new input somehow moves away from already formed attractor to another attractor. Pattern or process is called pattern separation.

pattern completion

two similar input patterns generate even more similar output patterns (for memory recall)

CA3 performs patterns completion and pattern separation

CA1 seems more linear

pattern separation

two different input patterns generate even more different output pattenrs (for new memory formation)

CA3 performs patterns completion and pattern separation

CA1 seems more linear

CA3 as a CPU for memory

how hebbian synaptic plasticity leads to memory?

learning: see face, visual neurons activated, have neurons in higher order, send input to enothrinal cortex, send info to distal dendrite of CA3 neuron, make neuron fire.

tell name, the sound send audtory neurons, to entorhinal, the next cell through synapse and another neuron will fire. the firirng of these neurons will generate synaptic plasticity

offline: because those two neurons activate together, the next synapse will also be potentiated. long term change in structure of synapses such that these connections will still be there during offline period

recall: next time when you see the face and memory will be recalled, in that process face neurons will be reactivated pass info to entorhinal cortex and will activate synapse that make specific neuron fire. because recurrent connection the firing of this neuron will activate synapse and make the other neuron fire. during recall even if you have one part of the OG info activated it can automatically recall the entire pathway and can recall name.

hippocampal input: entorhinal cortical grid cells

follow spikes of animal walking around randomly in space and plot all spikes in space. there are cluster spikes and plot the firing rate and see theres a regular pattern that can be computed as cross correlation.

grid firing location persist in the dark.

grid cells

Grid cells are specialized neurons in the brain's entorhinal cortex that function as an internal GPS, firing at regular intervals to create a hexagonal coordinate system for spatial navigation. They enable mammals to track location, distance, and direction.