lpetrich
Contributor
How Wrong Should You Be? - Scientific American Blog Network
In academic terms, the appropriate level of difficulty would be B or B+, even if they don't look as good on one's record as A's.
The Eighty Five Percent Rule for Optimal Learning | bioRxivMy best friend in college was a straight-A student, an English major. In part he got all A’s because he is whip smart—his essays were systematically better than everyone else’s. But the other reason was that he refused to enroll in a course unless he was certain he would ace it. Consequently, he never really challenged himself to try something beyond his comfort zone. I, on the other hand, was not a straight-A student. My first semester I took atomic physics with Professor Delroy Baugh, self-proclaimed “Laser Guy.” I’d never taken a physics course before in my life, and as a reward for my willingness to transcend my comfort zone I received a D.
Somewhere between the two of us lies a sweet spot: if you only ever get 100 percent on your tests, they aren’t hard enough. If you never get above 50 percent, you’re probably in the wrong major. The trick is to be right enough, but not so right that you never allow yourself the opportunity to be wrong.
So, how wrong should you be?
That number is from a theoretical argument that yielded (1/2) * (1 - erf(1/sqrt(2))) ~= 0.1587. It was tested with some numerical experiments in machine learning.Researchers and educators have long wrestled with the question of how best to teach their clients be they human, animal or machine. Here we focus on the role of a single variable, the difficulty of training, and examine its effect on the rate of learning. In many situations we find that there is a sweet spot in which training is neither too easy nor too hard, and where learning progresses most quickly. We derive conditions for this sweet spot for a broad class of learning algorithms in the context of binary classification tasks, in which ambiguous stimuli must be sorted into one of two classes. For all of these gradient-descent based learning algorithms we find that the optimal error rate for training is around 15.87% or, conversely, that the optimal training accuracy is about 85%. We demonstrate the efficacy of this 'Eighty Five Percent Rule' for artificial neural networks used in AI and biologically plausible neural networks thought to describe human and animal learning.
In academic terms, the appropriate level of difficulty would be B or B+, even if they don't look as good on one's record as A's.