Learning a new skill usually requires actively practicing it through trial and error while getting feedback on your performance. However, scientists have found that passive exposure to relevant information can also boost learning, even without direct training or feedback.
Researchers at UC Berkeley recently demonstrated this effect in mice learning to categorize sounds. They found that passive exposure to sounds before or during active training helped the mice learn faster. The scientists also built computer models simulating neural networks in the brain that showed how passive exposure could change how sensory information is represented and make it easier to link with the correct responses.
The results, published in Neuroscience, provide insights into more efficient training methods that combine passive and active learning. This hybrid approach could help learn real-world skills like musical instrument training or language acquisition in human adults and children.
The Key Ingredients: Effort, Practice, and Feedback
Learning a new perceptual skill requires actively making decisions about the task and getting feedback on those choices. Say you’re trying to better categorize sounds as high or low frequency. You would listen to tones, guess if they sounded high or low, and get told whether your choice was correct. This closed-loop choice, feedback, and adjustment process train the brain’s decision-making circuits.
Passively hearing sounds in the background shouldn’t help with this active learning process. You’re not making any decisions or getting told if you’re right or wrong. But humans and other animals are constantly exposed to sensory information daily, even while not deliberately practicing a skill. The UC Berkeley scientists wondered whether the brain might exploit this passive exposure opportunistically.
Past Research Hints At Passive Learning
The researchers were inspired by previous studies showing that early exposure to sounds can prime animals’ developing auditory systems. Baby rats reared in an environment with certain repeating sounds become better at discriminating similar sounds later.
In adult humans, some research also suggested that interleaving passive listening sessions with active training on auditory tasks could boost learning. However, less was known about how passive exposure interacts with active learning in mature animal models that could provide more biological insights.
The UC Berkeley team realized that machine learning might also offer some clues. Computer scientists have found that first pre-training neural network models on large unlabeled datasets can enable them to learn labeled tasks much faster with fewer training examples. The auditory system may employ a similar trick, first using passive exposure to unlabeled sounds to optimize its internal model of the world, thus jumpstarting task learning.
Mice Learn Faster After Passive Listening
The researchers set out to systematically test whether passive sound exposure enhances active auditory category learning using mice trained to detect sound patterns.
The mice had to classify short pitch-changing tones as rising or falling in frequency. They indicated their choice by poking their nose into the left or right port of a testing chamber. Over multiple daily training sessions, the mice got better and better at accurately categorizing the sounds.
The scientists compared groups of mice that received the standard training against two other groups. One group got additional passive sound exposure by playing background sounds in their home cages before active training. Another group of mice heard passive sounds interleaved throughout the active learning phase rather than just at the start.
In both cases, the mice exposed to extra passive sounds showed accelerated learning. They reached peak categorization performance faster than mice only getting active training. This demonstrated that passive sound exposure can enhance active auditory skill learning in mature mammals.
Neural Network Models Simulate Biological Learnings
To understand what mechanisms might allow passive exposure to confer learning benefits seen behaviorally, the researchers built computational models mimicking neural processing and learning in the auditory system.
They tested different architectures ranging from simple linear classifiers to multi-layer neural networks. Some models utilized only supervised learning, where network connections are tuned based on the match between model outputs and accurate labels during training. Other models were added in unsupervised pre-training, where patterns in the structure of unlabeled input data are captured before linking outputs to labels.
The researchers found the multi-layer neural networks that essentially did unsupervised feature detection from passive sound inputs followed by supervised fine-tuning best matched the mice’s learning performance. The models pointed to passive exposure leading to reorganized sensory representations that enable more efficient active learning.
Interleaving Works Best
In addition to showing that passive exposure accelerated learning, the experiments revealed that sound presentations interleaved throughout training led to faster mastery than passive listening only at the start. Further computational modeling helped explain this result through what’s known as congruency between the learning rules driving passive versus active model updates.
When happening together, the unsupervised and supervised shifts in neural connections compound each other, rapidly honing representations to enable superior performance. The researchers found just a handful of interleaved passive sessions to be as effective as days of initial passive exposure.
So when picking up that guitar or starting foreign language tapes, don’t just dive straight into hardcore practice immediately. Be sure to supplement your daily active training with passive listening. Your brain will soak up the sounds, and you’ll hit your learning goals in less time.
This approach combining passive exposure with active learning provides a recipe for faster mastery of real-world skills using our brains’ natural learning processes. Scientists studying how animals and AIs learn new tasks continue uncovering optimal training schedules that leverage labeled and unlabeled data. However, this research helps the biological or digital brain make the grade by revealing a way to boost learning.
About the Author
Robert Jennings is co-publisher of InnerSelf.com with his wife Marie T Russell. He attended the University of Florida, Southern Technical Institute, and the University of Central Florida with studies in real estate, urban development, finance, architectural engineering, and elementary education. He was a member of the US Marine Corps and The US Army having commanded a field artillery battery in Germany. He worked in real estate finance, construction and development for 25 years before starting InnerSelf.com in 1996.
InnerSelf is dedicated to sharing information that allows people to make educated and insightful choices in their personal life, for the good of the commons, and for the well-being of the planet. InnerSelf Magazine is in its 30+year of publication in either print (1984-1995) or online as InnerSelf.com. Please support our work.
Creative Commons 4.0
This article is licensed under a Creative Commons Attribution-Share Alike 4.0 License. Attribute the author Robert Jennings, InnerSelf.com. Link back to the article This article originally appeared on InnerSelf.com
books_perforamnce
