- "Overfitting" occurs when data points are too closely matched or repetitive; overfit data in machine learning makes it hard for AI to generalize.
- Much like overfitting curtails machine learning, the human brain may also be in danger of overfitting when daily training sets lack variety.
- A new AI-based dream theory, the Overfitted Brain Hypothesis (OBH), posits that dreaming may help the brain learn new things despite repetition.
As an ultra-distance triathlete who led a repetitious, monk-like existence for years on end, dreaming was an escape hatch from the monotony of daily life that also helped my brain master the motor skills required for peak performance.
For example, triathlon training sets for something like the Triple Ironman (7.2-mile swim, 336-mile bike, and 78.6-mile run) required about 6-9 hours per day of repetitive aerobic activity. By its very nature, training for extreme-distance triathlons and ultra-marathons is incredibly monotonous. In life and sport, boredom makes it hard to stay motivated and curious.
To sustain my motivation, I relied on my imagination and romanticized being a scrappy underdog. In my daydreams, I created mythical narratives that fueled my inspiration during tedious workouts and helped me cope with the daily grind. At night, I'd have the wildest dreams.
In my 2007 book, The Athlete's Way, I write about how dreaming helps athletes (and others who rely on finely-tuned muscle coordination) improve their performance based on the research of Robert Stickgold of Harvard Medical School's Division of Sleep Medicine.
In a July 2005 Harvard Magazine article, Stickgold explains how sleep and dreaming can improve motor skills:
"Suppose you are trying to learn a passage in a Chopin piano étude, and you just can't get it. You walk away, and the next day, the first try, you've got it perfectly. We see this with musicians and with gymnasts. There's something about learning motor-activity patterns, complex movements: they seem to get better by themselves, overnight."
In their recent (2021) book, When Brains Dream, co-authors Antonio Zadra and Robert Stickgold present a new dream theory they call network exploration to understand possibilities (NEXTUP). Hypothetically, NEXTUP allows dreamers to explore different narrative structures that may be impossible to pursue during wakefulness—and to consolidate memories in the process.
Six Contemporary Dream Theories
- Dreams are for memory consolidation
- Dreams are for emotion regulation
- Dreams are for selective forgetting
- Dreams are preparations for real-world problems
- Dreaming benefits predictive processing by refining generative models
- Dreaming opens our minds to unexplored possibilities (NEXTUP theory)
How Does the Overfitted Brain Hypothesis (OBH) Expand on Existing Dream Theories?
Erik Hoel of Tufts University published his overfitted brain hypothesis (Hoel, 2021) on May 14 in the peer-reviewed journal Patterns. This new AI-inspired dream theory posits that much like machine learning suffers from too much repetitive data, the human brain also struggles to learn new things when the distribution of daily stimuli includes too much of the same thing or becomes repetitious.
"Life is boring sometimes," he said in a news release. "Dreams are there to keep you from becoming too fitted to the model of the world." In his Patterns paper, Hoel writes, "Overtraining on a novel task creates the condition of the brain being overfitted to the task, which then triggers nightly dreams attempting to generalize performance on the task."
As an ultra-endurance triathlete who often overtrained, I realize now that dreaming may have helped my brain cope with overfitting caused by spending way too many hours per day doing the exact same training sets.
"The most effective means of triggering dreams that contain partial similarities to real-life events is through repetitive overtraining on a task," Hoel writes. "Put another way: The surest way to trigger dreams about a real-world event is to perform a task repetitively during the day, preferably one that is novel. Under the OBH, the explanation for this effect, as well as the specifics of the benefits of dreaming, is clearly outlined."
According to Hoel's new hypothesis, dreams may counteract overfitting by creating hallucinatory "out-of-distribution sensory stimulation every night." Theoretically, having really weird dreams makes it easier for the brain "to rescue the generalizability of its perceptual and cognitive abilities and increase task performance."
To counteract the eye-glazing and mind-numbing effect of a daily grind, Hoel posits that vivid dreams may help to highlight nuances in a sea of monotony. Strange dreams can alert the brain to something noteworthy that might otherwise go unnoticed in a repetitious training set. As he explains:
"The hallucinogenic, category-breaking, and fabulist quality of dreams means they are extremely different from the "training set" of the animal (i.e., their daily experiences). The [daily] cycle of fitting to tasks during the day, and avoiding overfitting at night via a semi-random walk of experiences, may be viewed as a kind of "simulated annealing" in the brain. That is, it is the very strangeness of dreams in their divergence from waking experience that gives them their biological function."
Hoel gives the example of a dream that involves flying possibly helping a runner stay balanced while running. As he explains:
"It may seem paradoxical, but a dream of flying may actually help you keep your balance running. The evidence for this possibility comes from common methods in deep learning that improve generalization, such as dropout, domain randomization, and the use of input data created by stochastic stimulation of generative models, which together bear striking similarities to the properties of dreams."
According to Hoel's overfitted brain hypothesis, dreaming may "provide departures away from the statistically biased input of an animal's daily life, which can assist and therefore increase performance."
Much Like Dreams, Art and Fiction May Also Offset Overfitting in the Human Brain
Interestingly, Hoel concludes his paper by stating: "Finally, it is worth taking the idea of dream substitutions seriously enough to consider whether fictions, like novels or films, act as artificial dreams, accomplishing at least some of the same function."
Anecdotally, as someone who regularly used daydreaming and fictional narratives to break up the monotony of training for ultra-endurance sports—the hypothesis that, in addition to nighttime dreams, putting yourself in the shoes of a fictional character prevents overfitting makes sense to me.
"The OBH suggests that fiction, and perhaps the arts in general, may actually have a deeper underlying cognitive utility in the form of improving generalization and preventing overfitting by acting as artificial dreams," Hoel concludes.
LinkedIn image: Sergey Nivens/Shutterstock. Facebook image: DedMityay/Shutterstock
Erik Hoel. "The Overfitted Brain: Dreams Evolved to Assist Generalization." Patterns (First published: May 14, 2021) DOI: 10.1016/j.patter.2021.100244