Humans' Deep Learning for Humans

Created by Maciej Sobczak

deep + learning (learnable) + human

learning?

human?

data = neural network

learnable?

All generalizations are false, including this one.

Mark Twain

the environment and generalization

The acquired skills should be useful also in the situations not encountered during the learning phase.

when generalization can occur?

• Invariance Assumption
• Learnable Regularity Assumption (detectable + detection algorithm is feasible)

animal vs plant

only certain (very limited) classes of functions can fulfil the above (linear separator and conjuctions).

• learning process takes limited number of steps
• the computation requires only limited number of interactions with the world
• learning leads to categorization with small error rate

teacher

evolution = learning?

why the world around us is learnable?

breeding the intelligent systems = neuroevolution

deep?

backpropagation

e = (2a + b)(a - b + 4)

deep representations?

can the representation be useful on its own?

image embeddings

word embeddings

human?

use case: foreign language acquisition

• hierarchically driven: phonemes, chunking, words
• naturally unfolding examples
• don't be lazy e.g. motor cortex used
• interest driven: fight with the attention selecting mechanism

Big big thank you to Christopher Olah for his blog and permission to adapt some of his examples for this talk

Just as there are odors that dogs can smell and we cannot, as well as sounds that dogs can hear and we cannot, so too there are wavelengths of light we cannot see and flavors we cannot taste. Why then, given our brains wired the way they are, does the remark “Perhaps there are thoughts we cannot think,” surprise you? Evolution, so far, may possibly have blocked us from being able to think in some directions; there could be unthinkable thoughts.

Richard Hamming

References

• Title slide background video build from: The Miracle in Human Brain