Track View Time View


  • Convolutional methods for Text in Tensorflow

    | Duration: 180min (Workshop)
    Bar-Ilan CS lab 202 (Building 604 on the map)

    RNNs are the default architecture for "deep NLP" but their are numerous practical cases where convolutions are as effective while training faster and requiring fewer computational resources.
    The first hour of this workshop will cover the convolutional operations we need to work with text including dilated convolutions, deconvolutions and tricks for managing gradients. We'll then use what we learnt to restore capitalization and punctuation to  text a common task when dealing with speech recognition. 
    Please come prepared with a laptop, tensorflow and nltk installed. Python 3 is preferred. The materials for the workshop are