̽��ֱ�� of Cambridge - Mate Lengyel

For the brain, context is key to new theory of movement and memory

Anonymous — Wed, 24 Nov 2021 16:00:00 +0000

How is it that a chef can control their knife to fillet a fish or peel a grape and can wield a cleaver just as efficiently as a paring knife? Even those of us less proficient in the kitchen learn to skilfully handle an astonishing number of different objects throughout our lives, from shoelaces to tennis rackets.

This ability to continuously acquire new skills, without forgetting or degrading old ones, comes naturally to humans but is a major challenge even for today’s most advanced artificial intelligence systems.

Now, scientists from the ̽��ֱ�� of Cambridge and Columbia ̽��ֱ�� (USA) have developed and experimentally verified a new mathematical theory that explains how the human brain achieves this feat. Called the COntextual INference (COIN) model, it suggests that identifying the current context is key to learning how to move our bodies.

̽��ֱ��model describes a mechanism in the brain that is constantly trying to figure out the current context. ̽��ֱ��theory suggests that these continuously changing beliefs about context determine how to use existing memories — and whether to form new ones. ̽��ֱ��results are reported in the journal Nature.

“Imagine playing tennis with a different racket than usual or switching from tennis to squash,” said co-senior author Dr Daniel Wolpert from Columbia ̽��ֱ��. “Our theory explores how your brain adjusts to these situations and whether to treat them as distinct contexts.”

According to the COIN model, the brain maintains a repertoire of motor memories, each associated with the context in which it was created, such as playing squash versus tennis. Even for a single swing of the racket, the brain can draw upon many memories, each in proportion to how much the brain believes it is currently in the context in which that memory was created.

This goes against the traditional view that only one memory is used at a time. To improve performance on the next swing, the brain also updates all memories, once again depending on its belief about the current context. When the context of the movement is judged to be new (the first time we play squash after years of tennis, for example), the brain automatically creates a new memory for that context. This ensures that we do not overwrite previously established memories, such as the memory for playing tennis.

This research may lead to better physical therapy strategies to help people with injuries use their bodies again. Often the improvements seen in the setting of a physical therapist's office do not transfer to improvements in the real world.

“With a better understanding of how context affects motor learning, you can think about how to nudge the brain to generalise what it learns to contexts outside of the physical therapy session,” said first author Dr James Heald. “A better understanding of the basic mechanisms that underlie the context dependence of memory and learning could have therapeutic consequences in this area.”

“What I find exciting is that the principles of the COIN model may also generalise to many other forms of learning and memory, not just memories underlying our movement,” said co-senior author Professor Máté Lengyel from Cambridge’s Department of Engineering. “For example, the spontaneous recurrence of seemingly forgotten memories, often triggered by a change in our surroundings, has been observed both in motor learning and in post-traumatic stress disorder.”

COINing a new model

Practice with a tennis racket, and the brain forms motor memories of how you moved your arm and the rest of your body that improve your serve over time. But learning isn’t as simple as just making better memories to make movements more precise, the researchers said. Otherwise, a tennis player’s serves might improve to the point at which they never hit a ball out of bounds. ̽��ֱ��real world and our nervous systems are complex, and the brain has to deal with a lot of variability.

How does the brain distinguish this noise — these random fluctuations — from new situations? And how does it understand that a slightly lighter tennis racket can still be operated using previous tennis racket memories? But that a table tennis paddle is an entirely different kind of object that requires starting from scratch?

̽��ֱ��answer, according to the COIN model, may be Bayesian inference, a mathematical technique used to deal with uncertainty. This method statistically weighs new evidence in light of prior experience in order to update one's beliefs in a changeable world. In the COIN model, a context is a simplifying assumption that, in a given set of circumstances, certain actions are more likely to lead to some consequences than others. ̽��ֱ��new theory's acceptance of the role that uncertainty plays in motor learning is similar to how quantum physics views the universe in terms of probabilities instead of certainties, the scientists noted.

Getting a handle on the theory

̽��ֱ��researchers put the COIN model to the test on data from previous experiments, as well as new experiments, in which volunteers interacted with a robotic handle. Participants learned to manipulate the handle to reach a target while the handle pushed back in different ways.

Volunteers who spent time learning to operate the handle as it pushed to the left, for instance, had more trouble operating the handle when it changed behaviour and pushed to the right, as compared to volunteers who started with a handle pushing to the right. ̽��ֱ��COIN model explained this effect, called anterograde interference.

“ ̽��ֱ��longer you learn one task, the less likely you are to move into a new context with the second task,” said Wolpert. “You’re still forming a motor memory of the second task, but you’re not using it yet because your brain is still stuck back in the first context.”

̽��ֱ��model also predicted that a learned skill can re-emerge even after subsequent training seems to have erased it. Called spontaneous recovery, this re-emergence is seen in many other forms of learning besides motor learning. For example, spontaneous recovery has been linked with challenges in treating post-traumatic stress disorder, where contexts can trigger traumatic memories to spontaneously recur.

Scientists usually explain spontaneous recovery by invoking two different learning mechanisms. In one, memories learned quickly are forgotten quickly, and in the other, memories learned slowly are forgotten slowly, and can thus reappear. In contrast, the COIN model suggests there is just one mechanism for learning instead of two separate ones, and that memories that apparently vanished may be ready to pop back with the right trigger: the belief that the context has re-emerged. ̽��ֱ��researchers confirmed this in their lab with new experiments.

Máté Lengyel is a Fellow of Churchill College. ̽��ֱ��research was supported by the European Research Council, the Wellcome Trust, the Royal Society, the National Institutes of Health, and the Engineering and Physical Sciences Research Council.

Reference:
James B Heald, Máté Lengyel and Daniel M Wolpert. ‘Contextual inference underlies the learning of sensorimotor repertoires.’ Nature (2021). DOI: 10.1038/s41586-021-04129-3

Adapted from a Columbia ̽��ֱ�� press release.

Mathematical model could help in physical therapy and shed light on learning more generally.

̽��ֱ��COIN model may also generalise to many other forms of learning and memory, not just memories underlying our movement

Máté Lengyel

Chino Rocha via Unsplash

Tennis match

̽��ֱ��text in this work is licensed under a Creative Commons Attribution 4.0 International License. Images, including our videos, are Copyright © ̽��ֱ�� of Cambridge and licensors/contributors as identified. All rights reserved. We make our image and video content available in a number of ways – as here, on our main website under its Terms and conditions, and on a range of channels including social media that permit your use and sharing of our content under their respective Terms.

Yes

Study identifies our ‘inner pickpocket’

sc604 — Tue, 21 May 2019 07:00:00 +0000

̽��ֱ��researchers, from the ̽��ֱ�� of Cambridge, the Central European ̽��ֱ��, and Columbia ̽��ֱ��, found that one of the reasons that successful pickpockets are so efficient is that they are able to identify objects they have never seen before just by touching them. Similarly, we are able to anticipate what an object in a shop window will feel like just by looking at it.

In both scenarios, we are relying on the brain’s ability to break up the continuous stream of information received by our sensory inputs into distinct chunks. ̽��ֱ��pickpocket is able to interpret the sequence of small depressions on their fingers as a series of well-defined objects in a pocket or handbag, while the shopper’s visual system is able to interpret photons as reflections of light from the objects in the window.

Our ability to extract distinct objects from cluttered scenes by touch or sight alone and accurately predict how they will feel based on how they look, or how they look based on how they feel, is critical to how we interact with the world.

By performing clever statistical analyses of previous experiences, the brain can immediately both identify objects without the need for clear-cut boundaries or other specialised cues, and predict unknown properties of new objects. ̽��ֱ��results are reported in the open-access journal eLife.

“We’re looking at how the brain takes in the continuous flow of information it receives and segments it into objects,” said Professor Máté Lengyel from Cambridge’s Department of Engineering, who co-led the research. “ ̽��ֱ��common view is that the brain receives specialised cues: such as edges or occlusions, about where one thing ends and another thing begins, but we’ve found that the brain is a really smart statistical machine: it looks for patterns and finds building blocks to construct objects.”

Lengyel and his colleagues designed scenes of several abstract shapes without visible boundaries between them, and asked participants to either observe the shapes on a screen or to ‘pull’ them apart along a tear line that passed either through or between the objects.

Participants were then tested on their ability to predict the visual (how familiar did real jigsaw pieces appear compared to abstract pieces constructed from the parts of two different pieces) and haptic properties of these jigsaw pieces (how hard would it be to physically pull apart new scenes in different directions).

̽��ֱ��researchers found that participants were able to form the correct mental model of the jigsaw pieces from either visual or haptic (touch) experience alone, and were able to immediately predict haptic properties from visual ones and vice versa.

“These results challenge classical views on how we extract and learn about objects in our environment,” said Lengyel. “Instead, we’ve show that general-purpose statistical computations known to operate in even the youngest infants are sufficiently powerful for achieving such cognitive feats. Notably, the participants in our study were not selected for being professional pickpockets -- so these results also suggest there is a secret, statistically savvy pickpocket in all of us.”

̽��ֱ��research was funded in part by the Wellcome Trust and the European Research Council.

Reference:
Gábor Lengyel et al. ‘Unimodal statistical learning produces multimodal object-like representations.’ eLife (2019). DOI: 10.7554/eLife.43942

Researchers have identified how the human brain is able to determine the properties of a particular object using purely statistical information: a result which suggests there is an ‘inner pickpocket’ in all of us.

These results suggest there is a secret, statistically savvy pickpocket in all of us

Máté Lengyel

Yes

Modelling how the brain makes complex decisions

sc604 — Thu, 04 Feb 2016 10:17:10 +0000

Researchers have constructed the first comprehensive model of how neurons in the brain behave when faced with a complex decision-making process, and how they adapt and learn from mistakes.

̽��ֱ��mathematical model, developed by researchers from the ̽��ֱ�� of Cambridge, is the first biologically realistic account of the process, and is able to predict not only behaviour, but also neural activity. ̽��ֱ��results, reported in ̽��ֱ��Journal of Neuroscience, could aid in the understanding of conditions from obsessive compulsive disorder and addiction to Parkinson’s disease.

̽��ֱ��model was compared to experimental data for a wide-ranging set of tasks, from simple binary choices to multistep sequential decision making. It accurately captures behavioural choice probabilities and predicts choice reversal in an experiment, a hallmark of complex decision making.

Our decisions may provide immediate gratification, but they can also have far-reaching consequences, which in turn depend on several other actions we have already made or will make in the future. ̽��ֱ��trouble that most of us have is how to take the potential long-term effects of a particular decision into account, so that we make the best choice.

There are two main types of decisions: habit-based and goal-based. An example of a habit-based decision would be a daily commute, which is generally the same every day. Just as certain websites are cached on a computer so that they load faster the next time they are visited, habits are formed by ‘caching’ certain behaviours so that they become virtually automatic.

An example of a goal-based decision would be a traffic accident or road closure on that same commute, forcing the adoption of a different route.

“A goal-based decision is much more complicated from a neurobiological point of view, because there are so many more variables – it involves exploring a branching set of possible future situations,” said the paper’s first author Dr Johannes Friedrich of Columbia ̽��ֱ��, who conducted the work while a postdoctoral researcher in Cambridge’s Department of Engineering. “If you think about a detour on your daily commute, you need to make a separate decision each time you reach an intersection.”

Habit-based decisions have been thoroughly studied by neuroscientists and are fairly well-understood in terms of how they work at a neural level. ̽��ֱ��mechanisms behind goal-based decisions however, remain elusive.

Now, Friedrich and Dr Máté Lengyel, also from Cambridge’s Department of Engineering, have built a biologically realistic solution to this computational problem. ̽��ֱ��researchers have shown mathematically how a network of neurons, when connected appropriately, can identify the best decision in a given situation and its future cumulative reward.

“Constructing these sorts of models is difficult because the model has to plan for all possible decisions at any given point in the process, and computations have to be performed in a biologically plausible manner,” said Friedrich. “But it’s an important part of figuring out how the brain works, since the ability to make decisions is such a core competence for both humans and animals.”

̽��ֱ��researchers also found that for making a goal-based decision, the synapses which connect the neurons together need to ‘embed’ the knowledge of how situations follow on from each other, depending on the actions that are chosen, and how they result in immediate reward.

Crucially, they were also able to show in the same model how synapses can adapt and re-shape themselves depending on what did or didn’t work previously, in the same way that it has been observed in human and animal subjects.

“By combining planning and learning into one coherent model, we’ve made what is probably the most comprehensive model of complex decision-making to date,” said Friedrich. “What I also find exciting is that figuring out how the brain may be doing it has already suggested us new algorithms that could be used in computers to solve similar tasks,” added Lengyel.

̽��ֱ��model could be used to aid in the understanding of a range of conditions. For instance, there is evidence for selective impairment in goal-directed behavioural control in patients with obsessive compulsive disorder, which forces them to rely instead on habits. Deep understanding of the underlying neural processes is important as impaired decision making has also been linked to suicide attempts, addiction and Parkinson's disease.

Reference:
Johannes Friedrich and Máté Lengyel. ‘Goal-Directed Decision Making with Spiking Neurons.’ ̽��ֱ��Journal of Neuroscience (2016). DOI: 10.1523/JNEUROSCI.2854-15.2016

Researchers have built the first biologically realistic mathematical model of how the brain plans and learns when faced with a complex decision-making process.

By combining planning and learning into one coherent model, we’ve made what is probably the most comprehensive model of complex decision-making to date

Johannes Friedrich

Seung Lab

EyeWire Candy Neurons

̽��ֱ��text in this work is licensed under a Creative Commons Attribution 4.0 International License. For image use please see separate credits above.

Yes

Licence type:

Attribution-Noncommercial-ShareAlike