Deep Learning For NLP

Organization

• Lecture: Dienstag 13:30-15:10, Room S101/A02
• Practice Class: Dienstag 15:20-17:00, Room S207/167
• Integrierte Lehrveranstaltung (TUCaN-ID 20-00-0947-iv)

The learning material is available from the Moodle eLeaning platform.

The required passcode will be distributed during the lecture.

The first lecture and the first practice class will be on April 10.

Course Content

The lecture provides an introduction to the foundational concepts of deep learning and their application to problems in the area of natural language processing (NLP)

Main aspects:

• Foundations of deep learning (e.g. feed-forward networks, hidden layers, backpropagation, activation functions, loss functions)
• Word embeddings: theory, different approaches and models, application as features for machine learning
• Different architectures of neuronal networks (e.g. recurrent NN, recursive NN, convolutional NN, encoder-decoder models) and their application for groups of NLP problems such as document classification (e.g. spam detection), sequence labeling (e.g. POS-tagging, Named Entity Recognition) and more complex structure prediction (e.g. Chunking, Parsing, Semantic Role Labeling)

Requirements:

foundations in math (linear algebra) and programming (python)

Expectations

What you can expect from us:

• interactive lecture with integrated tutorials
• problem-based and explorative learning
• stimulating environment

What we expect from you:

• commitment
• feedback
• active participation

Literature

• Yoav Goldberg (2015): A Primer on Neural Network Models for Natural Language Processing, published on arxiv
• Lecture by Richard Socher: Deep Learning for Natural Language Processing
• Ian Goodfellow, Yoshua Bengio, and Aaron Courville: Deep Learning, Book in preparation for MIT Press

Teaching Staff

• Steffen Eger (no fixed office hour, please just ask for an appointment per mail)
• Nils Reimers (no fixed office hour, please just ask for an appointment per mail)