How minds differ
Learning the detail, missing the pattern
In shortLearning tied too closely to one teacher, room or pencil may not show up anywhere else.
The idea
A good model of the world is simple enough to work in new situations. Overfitting is when learning sticks so closely to particular cases that it does not transfer: a skill learned with one teacher, in one room, with one pencil, may not appear with another. The opposite happens too: a model that is too broad (‘all dogs bite’, after one bite) ignores the differences that count. Learning well sits between the two.
What to try
- Practise the same skill in different places, with different people and materials.
- Point out what stays the same across examples: ‘different shops, the same way of paying’.
- After one bad experience, collect several good ones on purpose, so one case does not become the rule.
The common mistake
Concluding that a child ‘can’t’ when the skill has only ever been practised one way.
Want to know more?
Start here
Short, plain pieces by the researchers and good writers.
Books
How memory and thinking work, and what that changes in teaching and studying.
See on Amazon ›Amazon links are affiliate links: if you buy, Bounded Uncertainty receives a small commission at no cost to you.
The science behind itHow strong the evidence is, the model, videos and references.
How strong the evidence is
Promising Good studies, but little tested with children at home or school, or the step to daily life is our interpretation.
Variational free energy equals complexity minus accuracy, so minimising it penalises models that explain the data with more parameters than needed: a built-in guard against overfitting (Friston, 2010). FitzGerald, Dolan and Friston (2014) link model averaging and complexity to habit formation. Gershman and Niv (2010) describe learning as discovering latent causes: lumping too much underfits, splitting too much overfits. HIPPEA predicts overfitting through overweighted errors (Van de Cruys et al., 2014), and sleep appears to reduce model complexity (Hobson and Friston, 2012). In assessment: test generalisation explicitly by varying one element at a time (person, place, material, wording) and record where performance drops.
More technical books
The technical textbook on active inference, for those who want the formal model.
See on Amazon ›Amazon links are affiliate links: if you buy, Bounded Uncertainty receives a small commission at no cost to you.
References
- Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138.
- Gershman, S. J., & Niv, Y. (2010). Learning latent structure: carving nature at its joints. Current Opinion in Neurobiology, 20(2), 251–256.
- Hobson, J. A., & Friston, K. J. (2012). Waking and dreaming consciousness: Neurobiological and functional considerations. Progress in Neurobiology, 98(1), 82–98.
- FitzGerald, T. H. B., Dolan, R. J., & Friston, K. J. (2014). Model averaging, optimal inference, and habit formation. Frontiers in Human Neuroscience, 8, 457.
- Van de Cruys, S., et al. (2014). Precise minds in uncertain worlds: Predictive coding in autism. Psychological Review, 121(4), 649–675.
Found this useful? Buy Ricardo a coffee