Charla

Transfer Learning applied to NLP and Computer Vision

Pablo Vargas Ibarra, Manuel López Sheriff

Los humanos aprendemos. Cuando encontramos una solución buena a un problema, intentamos reutilizar lo aprendido para resolver desafíos similares. En el contexto del Deep Learning, podemos aplicar técnicas de transfer learning heredando arquitecturas de redes neuronales ya entrenadas o representaciones densas de palabras (Word Embeddings), como punto de partida en nuevos problemas. Este tipo de técnicas se han probado útiles en la práctica y es un campo de investigación muy activo actualmente. Nos ahorran tiempo y nos permiten obtener mejores resultados.

Introduciremos problemas tipo en los campos del Natural Language Processing y Computer Vision. Expondremos las dificultades, técnicas y computacionales, que presentan estos ejercicios. Y acudirá en nuestro auxilio de manera natural el transfer learning. Finalmente, os enseñaremos dónde encontrar los recursos necesarios y mostraremos algún ejemplo de cercana actualidad.

Medio Español Inteligencia Artificial Big Data / Data Science Ciencia / Investigación

Slides y material adicional

Viernes 15/03/2019

17:00 - 17:50

Track 4 (4.1.E03)

Sobre los ponentes

Pablo Vargas Ibarra

Kernel Analytics

Data scientist and mathematician passionate about the application of statistics, optimization and machine learning to business decision-making. Currently, working at Kernel Analytics, a boutique consultancy specialized in advanced analytics.

I like investing a great amount of my free time studying and trying new things about Data Science such as online courses and Kaggle competitions. Lastly, I am especially interested in Deep Learning, a boosting area in recent years.

Manuel López Sheriff

Kernel Analytics

I really enjoy learning mathematics and sharing knowledge.

Since I got in the Data Science world some years ago, I have been working in several projects of different nature (sport, telco, health...). My fundamental work tool is Python, and I continue building a powerful statistics basis.

Unstructured data (images, text, audio) shows itself as much more challenging topic once you know the basics of data analysis. That is why I am truly stimulated by Deep Learning. I have gone through a few courses and have worked in related projects, and would like to become an expert in the field.