Overview: This course introduces neural networks as well as specialised model structures to extract information from images and language data. These tools enable the researcher to transform high-dimensional and unstructured data such as historical maps, handwritten documents, aerial and satellite images into simpler representations that can be used as inputs to econometric models. The models are solved mathematically, implemented and optimised form scratch in TensorFlow or PyTorch. 

Sessions: Predictive modelling, Neural networks, Backpropagation, Better optimisation, Images as data, Convolutional networks, Language as data, Embedding networks, Recurrent networks, Transformers.
Coming soon: Auto-encoders, Bayesian networks, Graph networks.

Overview: Many important questions in development economics remain unanswered, partly because the data required to address them is encoded in high-dimensional data structures such as images. This course begins with an overview of novel data sources for development economists and then offers a comprehensive introduction to image processing techniques and specialised neural network models for analysing image data. While grounded in theory and mathematical formalisation, the course emphasises intuitive understanding and practical implementation using Python. By the end of the course, students will be able to leverage a variety of data sources for empirical work, including satellite and street-view images, historical maps, and documents, for tasks such as regression, classification, localisation (YOLO) and segmentation (UNet).

Sessions: Image processing, Predictive modelling, Neural networks, Convolutional networks, Learning representations.

Overview: With the recent digitisation of vast text corpora and significant advancements in natural language processing, development economists can now address a wide range of novel research questions. This course begins with an introduction to collecting, processing, and representing numerically text data, followed by an exploration of specialised deep learning techniques to model the semantic and structural structures of language. The course prioritises intuitive understanding and practical implementation using Python. By the end of the course, students will be able to utilise a variety of data sources for empirical work, including documents, tweets, speeches, and news transcripts, for tasks such as topic modelling, sentiment analysis, document similarity, named entity recognition, and more.

Sessions: Web-scrapping, Text processing, Word embeddings, Recurrent networks, Transformers.

Overview: This course introduces statistical learning for quantitative research in economics. Starting  with non-linear regression, the course provides a comprehensive understanding of some of the most capable supervised learning models such as random forest, gradient boosted trees and neural networks. Every model is solved mathematically before being implemented and optimised from scratch using Python.

Sessions: Predictive modelling, Generalised additive models, Support vector machines, Decision trees, Ensemble methods, Neural networks.

Overview: This course introduces regression methods for applied economic research. After covering multivariate regression and statistical inference, the course focuses on practical estimation issues such as collinearity, heteroscedasticity, serial correlation or endogeneity. Each problem is detected using the appropriate statistics and valid estimates are computed using alternative estimators. Sessions include comprehensive theorising yet keep a strong focus on intuition and effective implementation. We make extensive use the R programming language, both to illustrate abstract statistical concepts using simulated data, and to replicate research papers.

Sessions: Linear model, Inference, Specification , Model selection, Heteroscedasticity, Autocorrelation, Endogeneity, Identification.

• Lennart Schreiber, 2025, Panteon-Sorbonne University

• Olena Bogdan, 2022, ENS de Lyon

• Pablo Rodriguez, 2022, ENS de Lyon