Neuroscience News logo for mobile.

The brain applies data pressure to make a decision – Neuroscience News

Summary: Brain maximizes performance while minimizing cost by using data compression to help improve decision making.

source: Champalimaud Center for the Unknown

If you were a kid in the 80s, or a fan of retro video games, you must know Frogger. The game can be very challenging. To win, you must first survive a series of traffic jams, then narrowly escape oblivion by meandering through speeding wooden logs.

How does the brain know what to focus on in all this chaos?

A study published today in the scientific journal natural neuroscience Provides a possible solution: data compression.

“Compressing representations of the outside world is like removing all irrelevant information and adopting a temporary ‘tunnel view’ of the situation,” said study senior author Christian Machens, head of the Theoretical Neuroscience Laboratory at the Champalimaud Foundation in Portugal.

“The idea that the brain maximizes performance while minimizing cost using data compression is a prevalent idea in sensory processing studies. However, it has not really been examined in cognitive function,” said senior author Joe Patton, director of the Champalimaud Program for Neuroscience Research. “.

“Using a combination of experimental and computational techniques, we have shown that this same principle extends across a much wider range of functions than previously expected.”

In their experiments, the researchers used a timing model. In each trial, rats had to determine whether two tones were separated with an interval longer or shorter than 1.5 s. At the same time, the researchers recorded the activity of dopamine neurons in the animal’s brain as it performed the task.

“It is well known that dopamine neurons play a key role in learning the value of actions,” Machins explained. “So if the animal incorrectly estimates the duration of the time period in a given trial, the activity of these neurons will produce a ‘prediction error’ that will help improve performance in future experiments.”

Asma Motiwala, the study’s first author, built and tested a variety of computational reinforcement learning models that were better at capturing neuronal activity and animal behavior. The models shared some common principles, but differed in how they represented information that might be relevant to the performance of the task.

The team discovered that only models with a compressed task representation could compute the data. “It seems that the brain eliminates all irrelevant information. Oddly enough, it also apparently gets rid of some relevant information, but not enough to take a real hit on the amount of reward the animal collects in general,” said Machins. How do you succeed in this game?

Interestingly, the type of information provided was not only about the task variables themselves. Instead, it also captured the animal’s actions.

The models shared some common principles, but differed in how they represented information that might be relevant to the performance of the task. The image is in the public domain

Previous research focused on traits of the environment independently of an individual’s behavior. But we found that only compressed representations based on animal behavior fully accounted for the data.

“Indeed, our study is the first to show that the way representations of the outside world are learned, especially those that tax such a task, may interact in unusual ways with how animals choose to behave,” Motiwala explained.

According to the authors, this discovery has broad implications for neuroscience as well as for artificial intelligence. “While the brain has explicitly evolved to process information efficiently, AI algorithms often solve problems by brute force: using lots of data and lots of parameters. Our work provides a set of principles to guide future studies of how internal representations of the world support intelligent behavior in context,” Patton concluded. Biology and Artificial Intelligence.

About this research in Neuroscience News

author: Liad Hollander
source: Champalimaud Center for the Unknown
Contact: Liad Hollander – Champalimaud Center of the Unknown
picture: The image is in the public domain

see also

This shows a man sitting alone

original search: Access closed.
“Efficient coding of cognitive variables underlies dopamine responses and choice behaviour” by Afonso Vaz Pinto et al. natural neuroscience


Summary

Active coding of cognitive variables is the basis for dopaminergic responses and choice behaviour

Reward expectations based on internal knowledge of the external environment is an essential component of adaptive behavior. However, internal knowledge may be inaccurate or incomplete due to errors in sensory measurements. Some environment features may also be imprecisely coded to reduce representation costs associated with processing them.

In this study, we investigated how reward expectations are affected by the characteristics of internal representations by studying behavior and dopamine activity while rats make time-based decisions.

We show that several possible representations allow a reinforcement learning agent to model the animals’ overall performance during a task.

However, only a small subset of highly compressed representations simultaneously reproduced co-variance in animals’ selection behavior and dopaminergic activity. Remarkably, these representations predict an unusual distribution of response times that closely match the animals’ behaviour.

These results show how the limitations of representational efficiency can be expressed in coding representations of dynamic cognitive variables used in reward-based computations.

2022-06-06 18:09:05

Leave a Comment

Your email address will not be published. Required fields are marked *