Canada finally does something good

scruffy

Diamond Member
Mar 9, 2022
19,408
15,910
Canada funded this research.


For those of you following AI, this is big time.

Why it's important:

1. We already know there are grid cells, place cells, and time cells in the hippocampus of the brain. However before, there was only evidence of mapping to the external (perceptual) world.

2. We also already knew, that sensory images are encoded with a series of "hot spots" that map the most relevant feature locations. These hot spots are associated with criticality. Memory in the critical hot spots is 1000 times more efficient than in the rest of the network, which means a lot of detail can be encoded there.

3. What this research shows, is the hippocampus not only maps the external world, it maps the internal world as well. This way, hot spots in the sensory cortex becomes exactly like lights on the retina. In the same way the visual cortex is topographic to the retina, the hippocampus is topographic to locations in the brain. These locations are exactly the hot spots.

4. What this research suggests, is there are also grid, place, and time cells that map locations in the brain. This is a huge discovery, one of the most important in the last 30 years. Why? Because it tells how the egocentric reference frame is created. The area called "entorhinal cortex" is immediately adjacent to the hippocampus and is highly conserved in evolution. It hasn't changed much since goldfish. Layer 2 of the EC is mysterious, because it supports real time neurogenesis - and while other brain areas exhibit this as well, the EC generates replacements - 1500 cells die every day and 1509 more are born to replace them. Somehow, memory is transfered through the network into the new cells.

The hippocampus is necessary for transference of short term memory to long term memory, a process called consolidation. It is deeply affected in Alzheimer's. There is a famous patient known as HM who had a hippocampal lesion and couldn't remember anything for more than 30 minutes (although childhood memories were unaffected). Besides the EC there is an important pathway between the hippocampus and the lateral frontal lobe that is thought to provide "context" for short term memories. Without such context there can be no meaningful consolidation.

This research basically shows us what an engram looks like. Next we will study the behavior of the EC during reading, which is sequential and easily accessible by experiment.

Engineers will note that the grid and time cell mapping in the hippocampus uses "phase coding" relative to an external modulation, in this case a theta wave organized in the medial septal nucleus. The location of an event is mapped to the phase of the theta wave. So what we end up with is a "change of basis" from external topography to internal sequencing. This research shows how to get from A to B and back again.

 
Presumably it gives us a slight / better / closer understanding of how alien craft pilots use their brains / minds to instruct and perhaps fly their craft .
I wonder how we have progressed in this Beyond Black Ops area ?
Not that you would ever get an honest answer .
 
Presumably it gives us a slight / better / closer understanding of how alien craft pilots use their brains / minds to instruct and perhaps fly their craft .
I wonder how we have progressed in this Beyond Black Ops area ?
Not that you would ever get an honest answer .
1718359148509.png
 
Or the slightly more modern version being used to scan children with epilepsy

1718359395152.png


This one is actually interesting, it uses optically pumped magnetometers instead of the Josephson junctions normally used for MEG.

 
But back to topic: how to generate the egocentric reference frame:


Note the suggestion of "multiplexing", and the idea that this is done asynchronously using only local field potentials.

So basically you can have a cell coding at 6!Hz at the same time it's next door neighbor is coding at 11 Hz - and the kicker is you don't need the original modulation to recover the information.

Also there is a learning paradigm attached to this, called STDP (spike timing dependent plasticity). In the hippocampus it works like this: the first spike will travel in both directions away from the cell body. In the axon it signals downstream cells, but in the dendrites it causes a "phase change", (phase like gas liquid solid, instead of modulation), such that the dendrites stop signalling and instead modify their connection strengths. The time scale is 5-25 msec, the mechanism is dendritic calcium channels A hot spot has the fastest TTFS (time to first spike), so the area around the hot spot will rapidly encode local information even as the rest of the network is still operating normally. This is why we see hundreds of synapses between the same two neurons, like so:

1718360780141.png


The spike travels from one synapse to the next at about 1 m/sec, so if your synapse is 10 microns away that's about 10 microseconds or ,01 msec. This is why the hot spot is so efficient.

A single neuron will connect with about 10,000 other neurons, each of which has multiple synapses with slightly different timing. Done this way in the visual system a single hot spot can encode about 1/3 of the entire retinal image. Therefore it is not necessary to use all 10 million cortical neurons to store a visual image - only 3 neurons are needed. The eyes tremor at about 100 Hz, therefore the hot spots move around all the time. This it is possible to efficiently store moving visual images.
 

Forum List

Back
Top