Deep Learning for Computer Vision

Converting words to vectors

The words can be converted to vectors by training a model on a large text corpus. The model is trained such that given a word, the model can predict nearby words. The words are first one-hot encoded followed by hidden layers before predicting the one-hot encoding of nearby words. Training this way will create a compact representation of words. The context of the word can be obtained in two ways, as shown here: 

• Skip-gram: Given a single word, try to predict few words that are close to
• Continuous Bag Of Words (CBOW): Reverse of skip-gram by predicting a word given a group of words

The following image illustrates these processes:

Both methods show good results. The words are converted to vectors in an embedding ...