Spatial computing enables flexible working memory

Summary: The brain creates specific, distinct spaces in the cortex for each general rule of working memory and controls those spaces with brain rhythms, the researchers report.

Source: WITH

Routine tasks that require working memory, such as baking, involve remembering both some general rules (e.g. reading the oven temperature and time from the recipe and then setting them to the oven) and a specific content for each instance (for example, 350 degrees for 45 minutes for rye bread, but 325 degrees for 8 minutes for cookies).

A new study provides a new explanation for how the brain distinctly handles general and specific components of these cognitive demands.

Research by scientists from MIT’s Picower Institute for Learning and Memory and Karolinksa Institute and KTH Royal Institute of Technology in Stockholm, Sweden, shows that the brain creates distinct spaces in the cortex for each general rules and controls these patches with brain rhythms, a concept the authors call “spatial computing.”

This system, evident in animal study experiments, explains how the brain can easily maintain a consistent understanding of a process even when the specific content keeps changing (such as the time and temperature of bread versus cookies).

It also answers some questions that neuroscientists have pondered about the physiological operations that underlie working memory.

“Your brain can generalize instantly. If I teach you to follow certain rules, like memorizing C, A, and B and put them in alphabetical order, then change the content to F, D, and E, you won’t miss a beat,” said Earl K. Miller, Picower Professor at MIT’s Picower Institute for Learning and Memory and co-lead author of the study in Nature Communication.

“Your brain can do this because it represents rules and content at different physical scales. One can simply be plugged into the other.

How working memory works

Years of research by Miller’s lab, largely led by lead author Mikael Lundqvist who is now at Karolinska, have shown that working memory tasks are governed by an interaction of brain rhythms at distinct frequencies. The slower beta waves carry information about the rules of the task and selectively yield to the faster gamma waves when it is time to perform operations such as storing sense information or reading it out when recall is needed.

But these waves operate on networks of millions of neurons, only a handful of which actually store relevant individual information at any given time. Moreover, neurons that carry information about specific elements are found everywhere. Some become electrically aroused, or “spikes,” in response to different task rules than others, and they often tend to increase at least somewhat even when their information is irrelevant.

So how can these rather imprecise rhythms selectively control the right neurons at the right time to do the right things? Why are neurons whose peak relates to specific elements scattered and redundant? What makes a particular “350 degree” neuron straighten when that information needs to be stored, but another neuron with that information straightens when it needs to be recalled?

Researchers realized that all of these questions could be answered by spatial computation theory. Individual neurons representing pieces of information may be widely scattered around the cortex, but the rule applied to them is based on the patch of the network in which they are located. These patches are determined by the pattern of beta and gamma waves.

“Analyzing large numbers of unique neurons over the years, we always wondered why so many of them seemed to behave the same way,” Lundqvist said.

“Whether or not they prefer the same external stimulus, many neurons shared similar patterns of activity during working memory. And these diagrams passed from one task to another. It also appeared that neurons closer to each other in the prefrontal cortex more often shared the same pattern. This made us think that memory representations might actually flow dynamically through the prefrontal cortex to implement task rules.

Suppose your friend calls you at the gym to ask you to pick up a watch he accidentally left in his locker. This requires turning the dials of the padlock to the numbers of the combination (eg 24, 17, 32). Spatial computing says that when you hear the combination, your brain creates separate patches for each step (first, second, third).

In each patch, the neurons representing the combination number of that particular stage become particularly excited by the gamma waves applied at the time the rule is relevant (i.e. 24 in the “first” patch, 17 in the ” second” patch and 32 in the “third” patch).

In this way, individual neurons encoding specific information can be selectively associated with general rules by the brain waves controlling the areas they inhabit. In a given patch, all neurons may be somewhat excited by gamma waves, but those representing the element that matches the rule will be stimulated the most.

“In this way, memory representations could be dynamically reshaped to fit the demands of current tasks, regardless of how individual neurons are wired or what stimulus they prefer,” said the co- lead author Pawel Herman of KTH. “This may explain our impressive generalization abilities in new situations.”

This does not mean that a fix is ​​permanently fixed. The patches can come and go as long as they are needed wherever the brain is to train them for the task at hand. There is no permanent “remember oven temperature” patch in the brain.

“It gives flexibility to the brain,” Miller said. “Cognition is about flexibility.”

This shows the outline of a head in a ball held by a hand
Researchers realized that all of these questions could be answered by spatial computation theory. Image is in public domain

Experimental evidence

Researchers weren’t just theorizing. To test spatial computing in real physical brains, they made four experimental predictions about what they should observe while the animals played working memory games such as remembering a set of images in a order.

The first prediction was that there should be separate neural signals regarding rules and information about individual items. Indeed, the team measured that bursts of waves carried information about menstruation. The individual neural spikes, on the other hand, had a mix of individual items and task rules, consistent with them representing individual items and having specific rules imposed on them.

The second prediction was that rule information should be spatially organized and the third prediction was that these spatial patterns of rule enforcement should be consistent as long as the rules of the game remained the same, individual elements changed or not.

Sure enough, the researchers found that there were different locations for gamma-ray bursts for different rulers and that these remained stable even as the individual elements varied during each game.

The final prediction was that brainwave activity should cause neural spike activity to represent the right information at the right time. This was also reflected in experimental observations.

The researchers observed different brain wave patterns when the brain had to store images in memory and when it had to recall the “right” image. In general, beta waves were more reduced and neurons spiked more and in a wider area during recall than during storage.

The paper does not answer all questions about working memory. It is not yet clear how neurons encoding specific information in one patch might associate with their brethren in another patch or how the brain controls patches. Further research can answer these additional questions about the implications of new spatial computing theory.

About this memory and neuroscience research news

Author: Press office
Source: WITH
Contact: Press office – MIT
Picture: Image is in public domain

Original research: Free access.
“The Dynamics of Working Memory Control Follows the Principles of Spatial Computing” by Earl K. Miller et al. Nature Communication


The dynamics of working memory control follows the principles of spatial computing

Working memory (WM) allows us to memorize and selectively control a limited set of items. Neural evidence suggests that it is obtained by interactions between bursts of beta and gamma oscillations. However, it is unclear how oscillations, reflecting the coherent activity of millions of neurons, can selectively control individual WM elements.

Here we propose the novel concept of spatial computing where beta and gamma interactions cause element-specific activity to flow through space through the network during a task.

In this way, control-related information such as item order is stored in the spatial activity independently of the detailed recurrent connectivity supporting the item-specific activity itself.

Spatial flow in turn results in low-dimensional activity shared by many neurons. We verify these predictions by analyzing local field potentials and neural spikes.

We hypothesize that spatial computing can facilitate generalization and zero-hit learning by using the spatial component as an additional dimension of information encoding.

Leave a Comment