CBNS final exam (Models for Memory Storage

Recall Cajal, Konorski, Hebb

Cajal:

• Neuron doctrine: neurons communicate by contact not continuity (synapses)

Konorski

• associate learning as a result of long-term synaptic plasticity 

• “neural plasticity”

Hebb

• Hebbian Theory “neurons that wire together fire together”

• if cell A repeatedly helps fire cell B, some growth or metabolic change occurs in one or both neurons that increases A’s effectiveness in activating B

Hebbs cell assemblies

1. Before learning (all gray)

• Neurons exist but are not activated and not strongly connected.

• This full gray circle = the available network before any memory is formed.

• No part of it is a memory yet.

2. Stimulus arrives → initial activation (light red)

• A subset of neurons fires in response to the stimulus.

• Light-red = active but weakly connected neurons.

• This is the early internal representation of the stimulus.

3. Reverberating activity (still light red)

• Even after the external stimulus stops, the same neurons keep firing together briefly.

• This repeated co-activation sets up conditions for synaptic strengthening.

4. Hebbian modification → stronger connections (dark red appears)

• “Cells that fire together wire together.”

• The neurons that were repeatedly active strengthen their synapses.

• Dark-red = strongly connected neurons → the beginning of a memory trace (engram).

5. After learning → stable cell assembly (dark red cluster)

• The dark-red neurons together form the cell assembly.

• This assembly is the memory.

• Surrounding gray neurons = still inactive and not part of the memory.

**✔ The fully gray circle after Hebbian modification is just the rest of the network.
✔ The dark-red cluster inside it is the actual memory trace.

6. Partial cue → partial activation (light red again)

• Later, when only part of the original stimulus is presented, only a few of the original neurons activate (light red).

• These light-red neurons come from the partial reminder—not the full stimulus.

Examples:

• Seeing only part of an object

• Hearing only the first note of a song

• Smelling something faintly familiar

7. Pattern completion → full reactivation (dark red)

• Because the dark-red neurons are strongly interconnected, the activation spreads through the assembly.

• Light-red partial cue → triggers the entire dark-red cell assembly.

• This reactivation = retrieval of the memory.

A partial cue activates a subset of the original neurons (light red), which matches the stored engram (dark red) and triggers the entire strongly connected assembly to reactivate, resulting in full memory recall.

from Sg:

Internal representation

“All the cortical cells activated by an external stimulus.”
→ In the image, these are the light-red neurons that turn on when the stimulus appears.
This is the early activation pattern created by the stimulus.

Cell assembly

“A group of simultaneously active neurons.”
→ Those light-red neurons firing together = the cell assembly.
This is the temporary functional group that represents the stimulus.

Hebbian synapse / Autoassociation model

“A Hebbian synapse concept used in autoassociation computation.”
→ When the light-red neurons repeatedly fire together, their synapses strengthen.
→ This is shown as light red → dark red in the picture.
Autoassociation = the network linking itself together through Hebbian strengthening

Memory postulate

“Activity of the synapse + postsynaptic firing = synapse strengthens.”
→ This is literally Hebb’s rule (“fire together, wire together”).
In the diagram, this is when the dark-red, strongly connected pattern forms.
This is the engram (long-term memory trace)

Pattern completion

“If a partial input is presented, the autoassociation network completes the pattern.”
→ In the image, just a few neurons activate (light red),
→ and because of the strong dark-red connections, the whole pattern reactivates.
This is why seeing PART of a circle lets you recognize the WHOLE circle

Hebbs theory of consolidation

Step 1: Experience activates a pathway (yellow/light activation)

• Incoming sensory info activates certain neurons.

Step 2: Short-term memory — neurons fire together repeatedly

• The activated neurons keep firing in loops (reverberation).

• Think of this as the light red stage in the top image: active but not yet strengthened.

Step 3: Long-term memory — connections strengthen

• With continued co-activity, the synapses undergo structural + metabolic change.

• This corresponds to the dark red strong-connection stage from the first image.

• This strengthened path becomes a long-term memory trace

IN SUMMARY: Hebb and Konorski Cell assembly 

• External event activates cortical cells → inactive grey neurons become light red (active).

• Reciprocal connections → reverberation (active neurons keep firing together even after stimulus ends).

• Active neurons form a cell assembly (the group of light-red neurons firing in a pattern).

• Consolidation occurs through a “growth process” → repeated co-activity strengthens synapses.

• Synaptic strengthening shown as light red → dark red (high activity, strong conductivity).

• “Fire together, wire together” produces a stable engram (memory trace).

• This engram is widely distributed across the many linked cells within the assembly.

• Cell assemblies can include sensory and perceptual neurons, encoding features of the stimulus (e.g., “circle”).

recall 5 fundamental properties of LTP

• LTP is a prominent feature of the physiology of hippocampus, a brain structure universally identified with memory

• LTP develops rapidly (typically within 1 min.)

• LTP can be long lasting.

• LTP is input-specific. Only those synapse activated during the stimulation train are potentiated. Other neighboring synapses, even on the same neurons, are not altered.

• LTP is associative. The potentiation occurs best when multiple inputs are stimulated simultaneously during the tetanus. Note the relevance to Pavlovian conditioning, where CS and US are paired within few milliseconds of time window.

Model for Distibuted memory

Before Learning

• Neurons (A, B, C) respond weakly and similarly to all stimuli (Eric, Kyle, Kenny).

• No strong or distinct pattern yet.

Wrong Hypothesis: Nondistributed Memory

• Each memory stored in one neuron.

• Neuron A = Eric, Neuron B = Kyle, Neuron C = Kenny.

• Problem: biologically unrealistic and fragile — if one neuron dies, the memory is lost.

Correct Hypothesis: Distributed Memory

• Each memory is stored across many neurons, and each neuron participates in many memories.

• Recognition depends on the relative activation pattern across neurons.

• More resistant to damage; matches real neural network behavior.

Key Point

Human memory is distributed — not in single neurons, but in patterns of activation across a network.

Auto association network for recognition memory

• the strength of a particular synaptic connection is initially weak, but it will increase if the presynaptic and postsynaptic neurons are active at the same time

• This circuit framework with an embedded synaptic “learning rule”====AUTOASSOCIATION

• Autoassociation: the output is fed back to the input, where it associates itself.