The Unreasonable Effectiveness of Deep Feature Extraction
http://www.basilica.ai/blog/the-unreasonable-effectiveness-of-deep-feature-extraction/ [www.basilica.ai]
2019-02-15 04:39
A feature extractor is anything that takes in data and spits out a set of numbers that describe the important parts of it, in the same way a tailor might describe your shape with a small set of measurements.
Once they had this feature extractor, they fed images from smaller datasets into it, and then fed the resulting features into an support-vector machine, a very simple model that’s existed since the 90s. Basically, the OverFeat network was used to preprocess images so that they could be modeled using well-known ML tools, like a mother bird regurgitating partially-digested food for her young.
Such a lovely analogy.
source: HN