Perception is mostly about expectations. You can only perceive what you expect.
Even a newborn makes fundamental assumptions about the structure of its sensorimotor space (priors), without which it simply couldn't start making sense of (& learning from) its sensorimotor feed.
Learning, both at the individual level and at the collective level (e.g. science) is largely about expecting something (formulating a hypothesis and running an experiment to test it) then checking whether reality validates or contradicts the expectation. To learn is to expect.
The machine learning angle here is that you can only learn within your hypothesis space (search space), which can never contain "everything", if only because you have to encode it (describe it) somehow. To learn from data, you have to make assumptions about it.
The stronger the assumptions the faster you learn (e.g. densely connected network vs. convnets for spatial perception).
You can follow @fchollet.
Tip: mention @threader_app on a Twitter thread with the keyword “compile” to get a link to it.
Enjoy Threader? Sign up.
Threader is an independent project created by only two developers. The site gets 500,000+ visits a month and our iOS Twitter client was featured as an App of the Day by Apple. Running this space is expensive and time consuming. If you find Threader useful, please consider supporting us to make it a sustainable project.