Do you want to learn how to make use of text data for your research?
This in-person workshop is ideal for researchers who want to learn the potential of Natural Language Processing (NLP) and Large Language Models (LLMs) for economic and related research.
You will receive a broad overview of the theoretical foundations of NLP and its practical concepts. The workshop closes with a Coding Lab to apply this knowledge to real world problems.
Prerequisites
A basic understanding of Machine Learning and Deep Learning is required.
Basic Python skills are required. Familiarity with common modules for text processing and deep learning frameworks is recommended.
Schedule
09:30 – 11:00 NLP Basics (1-3)
11:15 – 12:45 Neural Nets & Transfer Learning (4-8)
13:45 – 15:45 Generative Models (9-End)
16:00 – 17:00 Coding Lab
Topics
1 Learning Paradigms
Understand the different learning paradigms, Relate type of learning to amount of labeled data required
2 NLP tasks
Understand the different types of tasks (low- vs. high-level), Purely Linguistic tasks vs. more general classification tasks
3 Word Embeddings
Understand what word embeddigns are, Learn the main methods for creating them
4 Recurrent Neural Networks
Understand recurrent structure of RNNs, Learn the different types of RNNs
5 Attention
Understand attention mechanism, Learn the different types of attention, The Transformer / Self-Attention
6 The BERT Architecture
Use of the transformer encoder in this model, Understand the pre-training, Gain understanding of the fine-tuning procedure Differences between token- and sequence classification
7 BERTology
Understand how impactful this architecture was, See how this changed research in the field, Glimpse into BERTology
8 Model distillation
soft vs. hard targets, understand how distillation works, DistilBERT, other approaches towards compression
9 Towards a unified task format
developments of the post-BERT era, reformulating classification tasks, multi-task learning, fine-tuning on task-prefixes
10 GPT series
use of the transformer decoder, input modifications (and how this is useful), concept of prompting