Summary: New research reveals that the brain may be learning even during unstructured, aimless exploration. By recording activity in tens of thousands of neurons, scientists found that the visual cortex builds internal models of the environment, preparing the brain for future tasks.
This unsupervised learning occurs without any instruction, helping animals learn goal-oriented tasks faster later on. The study highlights how both unsupervised and supervised learning operate in parallel within the brain, reshaping our understanding of how we acquire knowledge.
Key Facts:
- Unsupervised Learning: The brain encodes environmental features even without tasks, aiding future learning.
- Visual Cortex Roles: Distinct regions of the visual cortex handle exploratory (unsupervised) and task-based (supervised) learning.
- Faster Task Learning: Mice exposed to unstructured environments learned reward-linked tasks more quickly than unexposed peers.
Source: HHMI
Aimlessly wandering around a city or exploring the new mall may seem unproductive, but new research from HHMI’s Janelia Research Campus suggests it could play an important role in how our brains learn.
By simultaneously recording the activity of tens of thousands of neurons, a team of scientists from the Pachitariu and Stringer labs discovered that learning may occur even when there are no specific tasks or goals involved.
The new research finds that as animals explore their environment, neurons in the visual cortex—the brain area responsible for processing visual information—encode visual features to build an internal model of the world. This information can speed up learning when a more concrete task arises.
“Even when you are zoning out or just walking around or you don’t think you are doing anything special or hard, your brain is probably still working hard to help you memorize where you are, organizing the world around you, so that when you’re not zoning out anymore—when you actually need to do something and pay attention—you’re ready to do your best,” says Janelia Group Leader Marius Pachitariu.
Observing unsupervised learning
The team, led by postdoc Lin Zhong, designed experiments where mice ran in linear virtual reality corridors featuring various visual textures, akin to real-world environments. Some textures were linked to rewards, while others were not. After the mice learned the rules of an experiment, Zhong made subtle adjustments, altering the textures and the presence of rewards.
After weeks of running these experiments, the team observed changes in neural activity within the animals’ visual cortex. However, they struggled to explain the observed neural plasticity—the changes in connections between neurons that enable learning and memory.
“As we thought more and more about it, we eventually ended up on the question of whether the task itself was even necessary,” Pachitariu says. “It’s entirely possible that a lot of the plasticity happens just basically with the animal’s own exploration of the environment.”
When the researchers explicitly tested this concept of unsupervised learning, they discovered that certain areas of the visual cortex were encoding visual features even without the animal being trained on a task. When a task was introduced, other areas of the cortex responded.
Additionally, the researchers found that mice exploring the virtual corridor for several weeks learned how to associate textures with rewards much faster than mice trained only on the task.
“It means that you don’t always need a teacher to teach you: You can still learn about your environment unconsciously, and this kind of learning can prepare you for the future,” Zhong says.
“I was very surprised. I have been doing behavioral experiments since my PhD, and I never expected that without training mice to do a task, you will find the same neuroplasticity.”
Understanding how brains learn
The new findings reveal distinct areas in the visual cortex are responsible for different types of learning: unstructured, exploration-based unsupervised learning and instructed, goal-oriented supervised learning.
The new research suggests that when animals learn a task, the brain might simultaneously use both algorithms—an unsupervised component to extract features and a supervised component to assign meaning to those features.
These insights could enhance our understanding of how learning occurs in the brain. While previous research on the visual cortex focused mainly on supervised learning, the new work opens new avenues for exploration, including how these different types of learning interact and how the visual model of the environment is integrated with spatial models from other brain regions.
“It’s a door to studying these unsupervised learning algorithms in the brain, and if that’s the main way by which the brain learns, as opposed to a more instructed, goal-directed way, then we need to study that part as well,” Pachitariu says.
The researchers say these insights were enabled both by Janelia’s support teams, which helped the researchers design and run the experiments, and by the mesoscope, an instrument that enabled the team to record up to 90,000 neurons simultaneously, enhancing their ability to make new discoveries.
“Allowing a single lab to run projects at this scale is what is uniquely possible here and that gives us the flexibility to pursue different questions without necessarily having a concrete plan,” Pachitariu says.
About this neuroscience research news
Author: Nanci Bompey
Source: HHMI
Contact: Nanci Bompey – HHMI
Image: The image is credited to Neuroscience News
Original Research: Open access.
“Unsupervised pretraining in biological neural networks” by Marius Pachitariu et al. Nature
Abstract
Unsupervised pretraining in biological neural networks
Representation learning in neural networks may be implemented with supervised or unsupervised algorithms, distinguished by the availability of instruction.
In the sensory cortex, perceptual learning drives neural plasticity, but it is not known whether this is due to supervised or unsupervised learning.
Here we recorded populations of up to 90,000 neurons simultaneously from the primary visual cortex (V1) and higher visual areas (HVAs) while mice learned multiple tasks, as well as during unrewarded exposure to the same stimuli.
Similar to previous studies, we found that neural changes in task mice were correlated with their behavioural learning.
However, the neural changes were mostly replicated in mice with unrewarded exposure, suggesting that the changes were in fact due to unsupervised learning.
The neural plasticity was highest in the medial HVAs and obeyed visual, rather than spatial, learning rules. In task mice only, we found a ramping reward-prediction signal in anterior HVAs, potentially involved in supervised learning.
Our neural results predict that unsupervised learning may accelerate subsequent task learning, a prediction that we validated with behavioural experiments.
