Deep Learning in NLP
Asynchronous Online Course
- Length: 10 weeks (20 hrs)
- A “Certificate of Completion” will be awarded to you, if you pass the final exam.
If you know the basic underlying of machine learning and you would like to learn how to run an AI project on time-series data, this is the right place for you to learn the state-of-the-art sequence models. More specifically, we discuss the natural language processing (NLP) in which data is a sequence of words/letters.
In the first part, we learn the basics of NLP and recurrent neural networks (RNNs) which are widely used in sequence modeling for time-series data. Besides a multi-label classification task in Lab 1 with Python, we experience a deep learning implementation for sequence tagging in Lab 2 with PyTorch DL framework. The second part also discusses attention mechanisms and Transformers, which are also state-of-the-art sequence modeling techniques. We learn how to use pre-trained models in a transfer learning fashion for new tasks with small supervised data.
Avoiding from the complex mathematics, this course comprehensibly presents the underlying concepts of the deep learning models for time-series data (RNNs) and their applications. Moreover, we will review attention models. The concepts are supported by Lab session(s) by introducing popular python libraries and implementing some samples.
- Introduction (Definitions, Applications, Tasks, Approaches)
- Basics of Linguistics (NLP Pyramid, Vectorizations: BoW, TF-IDF, word2vec)
- Lab 1. Automatic Tagging in Python
- Recurrent Neural Networks (ML Components, LSTM, GRU)
- Lab 2. Named Entity Recognition in PyTorch
- Attention mechanisms & Transformers
- Lab 3. Spam classification with BERT in PyTorch
- Chatbots & Question/Answering
- Lab 4. Q/A with BERT in PyTorch
- It is highly suggested to complete the Introduction to AI & ML and Unsupervised Learning courses in advance.
- Basic Python programming is needed. For the lab session, we use Google Colab. So, you don’t need to install anything in your machine. The DL programs employs PyTorch framework.
The AI & ML Series
DL for NLP with Python is the third course in the AI & ML series. List of courses in the AI & ML boot camp is as follows:
Section 1. Introduction to NLP
Natural Language Processing (NLP) is an interdisciplinary field of computer science, AI, and linguistics. We will learn NLP applications, tasks, and approaches in this section.
Section 2. Basics of Linguistics
We will learn basics of linguistics, including NLP Pyramid, current challenges, and vectorization techniques in this section.
Lab 1. Automatic Tagging in Python
Automatic taggong on StackOverflow dataset is a multi-label classification task. We review the implementation in Python here.
Section 4. Recurrent Neural Networks
Recurrent Neural Networks (RNNs) are widely used in time-series analysis for sequence modeling. RNNs are utilized in NLP as text is also a type of sequence.
Lab 2. NER with LSTM in PyTorch
We use LSTM for seqence tagging, specifically for Named Entity Recognition. We learn how to implemnent the RNN in PyTorch in this lab section.
Section 6. Transformers
Transformer is a successful neural network architecture, useful for natural language processing. We will learn the concept of the attention mechanism; then we review the transformer architecture.
Section 7. BERT
BERT is a promising language model, applied in many NLP applications. It employs the encoder part of the transformer. We will learn the BERT architecture; then, we learn how to use the pre-trained BERT model.
Lab 3. Spam Classification with BERT in PyTorch
As a simple example, in this lab session, we use the pre-trained BERT model for spam classification. The code is in PyTorch.
Section 9. Chatbots & Question/Answering Models
Chatbots are widely used for entertainment and marketing. We review different types of chatbots.
Lab 4. Q/A with BERT in PyTorch
The Google QUEST Kaggle competition is a Q/A labeling task, required for Q/A chatbots. We review the code for the winner of the competition.
0 % In Progress
0 % In Progress
93.94 % Passed
93.94 % Passed
96.97 % Passed