Interest & curiosity
Interestingness is the first derivative of the beauty :
… the extent to which data will improve the compressor (world model) for observer .
Curiosity is the drive to improve the compressor:
Interesting data is usually novel but needs to also be learnable, in other words: open-ended
Predictable/unsurprising data (e.g. black room) → no learning progress, not interesting
Unable to extract patterns from data (e.g. white noise) → unlearnable, not interesting
Data that improves your ability to model the world → learning progress, interestingYou seek out information that leads to (new) higher level concepts, connections etc., exploring parts of the space where you know that your world model is inaccurate.
self-modelling]] is necessary for curiosity. If the learning algorithm depends on the model network, the model network has to make a prediction about its own current prediction capabilities. The activations of the model network are (partly) interpreted as a statement about the current weights of the model network.
Transclude of A-possibility-for-implementing-curiosity-and-boredom-in-model-building-neural-controllers#^7328c9
Circular transclusion detected: A-possibility-for-implementing-curiosity-and-boredom-in-model-building-neural-controllers
The gratification of curiosity rather frees us from uneasiness than confers pleasure; we are more pained by ignorance than delighted by instruction. (Johnson, 1751)
Embracing curiosity eliminates the exploration-exploitation dilemma
Todo
Curiosity-driven Exploration by Self-supervised Prediction
https://people.idsia.ch/~juergen/artificial-curiosity-since-1990.html
https://www.semanticscholar.org/paper/A-possibility-for-implementing-curiosity-and-in-Schmidhuber/2980dfe5c99658dc3e508d9d6e1d7f26e6fc8934#citing-papers read this paper + go through the citations
Large-Scale Study of Curiosity-Driven Learning