Information-theoretical label embeddings for large-scale image classification

@article{Chollet2016InformationtheoreticalLE,
  title={Information-theoretical label embeddings for large-scale image classification},
  author={François Chollet},
  journal={CoRR},
  year={2016},
  volume={abs/1607.05691}
}
We present a method for training multi-label, massively multi-class image classification models, that is faster and more accurate than supervision via a sigmoid cross-entropy loss (logistic regression). Our method consists in embedding high-dimensional sparse labels onto a lower-dimensional dense sphere of unit-normed vectors, and treating the classification problem as a cosine proximity regression problem on this sphere. We test our method on a dataset of 300 million high-resolution images… CONTINUE READING