...when an AI system becomes too familiar with data, it can oversimplify its analysis, becoming an "overfitted brain" that assumes what it sees is a perfect representation of what it will encounter in future. To counter that problem, scientists introduce a degree of chaos and randomization into their data to deepen machine learning and improve the accuracy of AI systems.
In much the same way, "our brains are so good at learning that we're always in danger of being overfitted,"— Medscape
warns Neuroscientist Erik Hoel, PhD, from Tufts University in Medford, Massachusetts, saying that this over-familiarity with data “can lead to overly simplistic and too-familiar perceptions of the world around us.”
He theorizes that our brains combat this hazard by introducing chaos into sleep, generating outlandish dreams to help our brains grasp day-to-day experiences in a deeper learning way than otherwise would happen. ... that the brain is more malleable and capable than previously thought.
From his recent study report (that’s free full text at cell.com):
The bigger picture
Dreaming remains a mystery to neuroscience. While various hypotheses of why brains evolved nightly dreaming have been put forward, many of these are contradicted by the sparse, hallucinatory, and narrative nature of dreams, a nature that seems to lack any particular function. Recently, research on artificial neural networks has shown that during learning, such networks face a ubiquitous problem: that of overfitting to a particular dataset, which leads to failures in generalization and therefore performance on novel datasets. … The overfitted brain hypothesis is that the brains of organisms similarly face the challenge of fitting too well to their daily distribution of stimuli, causing overfitting and poor generalization. By hallucinating out-of-distribution sensory stimulation every night, the brain is able to rescue the generalizability of its perceptual and cognitive abilities and increase task performance.
According to the Medscape article, although it has become routine for neuroscientists to examine neural activity during sleep, capturing dreams in order to attempt evaluation presents … challenges. Yet even if direct study of dreams is beyond current technology, Hoel thinks neuroscience has gotten closer, and he suggests that a
reliable way to integrate real-life experience into a dream is to perform a new task repetitively before sleep ... That process [then] triggers overfitting and, during sleep, the brain will create quirky dreams that incorporate the newly acquired knowledge.
Interesting, huh? Commenters who try this, please come back tomorrow and report on how it worked out.
;-)
RELATED <small>(also @ Medscape, no paywall but reg’n required)</small>:
Experimental data suggest that sleep is required for clearance of waste products from brain metabolism. This has, however, never been verified in humans. The primary aim of the present study was to examine in vivo whether one night of total sleep deprivation affects molecular clearance from the human brain. Secondarily, we examined whether clearance was affected by subsequent sleep. Multiphase MRI with standardized T1 sequences was performed up to 48 h after intrathecal administration of the contrast agent gadobutrol (0.5 ml of 1 mmol/ml), which served as a tracer molecule. Using FreeSurfer software, we quantified tracer enrichment within 85 brain regions as percentage change from baseline of normalized T1 signals. The cerebral tracer enrichment was compared between two cohorts of individuals; one cohort (n = 7) underwent total sleep deprivation from Day 1 to Day 2 (sleep deprivation group) while an age and gender-matched control group (n = 17; sleep group) was allowed free sleep from Day 1 to Day 2. From Day 2 to 3 all individuals were allowed free sleep. The tracer enriched the brains of the two groups similarly. Sleep deprivation was the sole intervention. One night of sleep deprivation impaired clearance of the tracer substance from most brain regions, including the cerebral cortex, white matter and limbic structures, as demonstrated on the morning of Day 2 after intervention (sleep deprivation/sleep). Moreover, the impaired cerebral clearance in the sleep deprivation group was not compensated by subsequent sleep from Day 2 to 3. The present results provide in vivo evidence that one night of total sleep deprivation impairs molecular clearance from the human brain, and that humans do not catch up on lost sleep.
Early morning exposure to blue wavelength light can help heal the brain following mild traumatic brain injury (mTBI), new research suggests.
Results of a small, randomized controlled trial showed blue-light therapy improved brain structure and function, cognition, and sleep in this patient population … "Prior research had shown that if you use blue light in people with mild traumatic brain injuries, it makes them feel less fatigued during the day, but it had not shown the physical changes in the brain that we demonstrated.
"What we're seeing now is that blue light is changing the actual brain structure by increasing volume in certain areas and increasing myelination of important pathways involved in visual attention..."
I like medscape. :) Professional level, highly readable stuff, free to everyone, just hafta register.