Introducción a Deep Learning

Session 1: basic introduction to supervised learning. definition of loss function, derivation of logistic regression, SGD. Focus on the importance of the representation. Vector Space Model (Salton). Feature extraction for NLP. application to toy problem (NER without structured prediction?, synonym prediction?, TBD)

Session 2: Hands-on. Code logistic regression from scratch, in sklearn format (transform_fit & predict_proba). Code some basic features extractors (using sklearn.pipeline)

Session 3: distributional representation of words. Firth 1963, Harris 1954. co-occurence matrix, point-wise mutual information. Bengio et al 2003, Mikolov et al 2013, Pennington et al 2014, Levy et al 2015.

Session 4: advanced topics. supervised-fine tuning of word embeddings, the problem of homonyms, counter-fitting word-embedings (using prior knowledge), multi-lingual embedddings

Session 5: Hands-on. Change feature extractors to embeddings (gensim library).

Desarrollo
lunes 4
13:30 - 15:30 Introducción a Deep Learning -- Matthias Gallé
15:30 - 17:30 Aprendizaje con pocos ejemplos -- Jorge Sánchez
martes 5
13:30 - 15:30 Introducción a Deep Learning -- Matthias Gallé
15:30 - 17:30 Tagging y Parsing en Procesamiento del Lenguaje Natural -- Franco Luque
miércoles 6
13:30 - 15:30 Introducción a Deep Learning -- Matthias Gallé
15:30 - 17:30 Express Deep Learning con Python -- Cristian Cardellino y Milagro Teruel
jueves 7
13:30 - 15:30 Introducción a Deep Learning -- Matthias Gallé
15:30 - 17:30 Data Mining en Redes Sociales con Python -- Pablo Celayes y Gabriel Miretti
viernes 8
13:30 - 15:30 Introducción a Deep Learning -- Matthias Gallé
15:30 - 17:30 Planning -- Carlos Areces

Coordinacion:

Laura Alonso Alemany (FaMAF-UNC)