You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
This particular upside-down facial feature perception glitch is known as the Thatcher effect.
Compositionality is also very salient in children's drawings. On that note, children's drawings are an amazing source of insights about the human mind.
What is your opinion on extending the idea of reference systems (the template above) beyond spatial perception, as in Jeff Hawkins latest theories?
Does anyone ever amplify the training set size by rotating, stretching & skewing images? Also, could you rotate, stretch & skew an image attempting to be classified, producing a classification for each transform, then just take the most common classification?
Seems like you could "wiggle" the input data a little bit, and see how varied the classification results are. If the classification doesnt change, thats more confidence than if each wiggle results in a different classification
@readwiseio save thread
First public save of this thread! 
Readwise users: Like this reply to save fchollet's thread to your account without cluttering their replies 
Stats:
• 73 total saves of fchollet's threads (ranked #441)
Its amazing that kids perceive and model things as stick figures while Conv networks seem to discriminate mostly based on the textures and need insane amount of data augmentation to come close to choosing boundaries as significant features.
That said, a moving 3d world can naturally provide a almost infinite set of data augmentations. Combined with other sensory modalities this dataset can be richer than anything we can manufacture.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
