Advanced Deep Learning for NLP
Teacher: Cristiano De Nobili – Harman-Samsung
This lecture is intended to be an advanced Deep Learning lecture on NLP.
In the first part, we touch some relevant concepts in NLP, such as word and character embedding. In addition, we review with the ‘eye of a physicist‘ a few Information Theory quantities which are fundamental in machine learning.
During the second part, we understand how to build a Seq2seq (encoder/decoder) algorithm and how to train it. This architecture is at the core of many state-of-the-art NLP applications, such as language translators.
Case study: we will build a spell checker which is able to correct spelling mistakes in a sentence.
This is a simple model which takes advantage of the attention mechanism. We will also devote some time to build our dataset. This exercise is also thought to be an opportunity to test and learn the latest versions of TensorFlow (1.13 and 2.0).
Prerequisites: Curiosity (a lot), Python (a bit), TensorFlow (a bit), and Deep Learning basics.
#deeplearning #NLP #tensorflow #python #machinelearning