‘Fast Convex Iteration Learning: Theory and Applications’

Álvaro Jiménez Barbero, Julia Díaz García, David Díaz Vico, L Jorge López Lázaro, Instituto de Ingeniería del Conocimiento; and Ana González Marcos and Alberto Torres Barrán, Universidad Autónoma de Madrid.

The project Fast Convex Iteration Learning, FACIL, aims first to perform basic research on convex optimization, seeking to contribute to the state of the art in the area of acceleration methods to improve the convergence of gradient descent in convex problems, with Support Vector Machines (SVMs) and sparse models (Lasso, Total Variation) as very important examples.

The project`s second goal is to apply some of these advances as well as other complementary Machine Learning approaches, particularly Deep Neural Networks, to achieve a better prediction of wind and solar energy, a field of interest on itself and also because of the relevant role that Spain plays in the renewable industry.

Because of this, the project is organized into two parts:

1. Better descent directions for SVMs and sparse linear models.
2. Sparse models and deep networks in wind and solar energy prediction.

A key project component is the partnership with the Instituto de Ingeniería del Conocimiento (IIC), an innovation center with more than 25 years of activity. IIC has its premises at the Escuela Politécnica Superior of the UAM and has the UAM as one of its sponsors, together with IBM Spain, Gas Natural Fenosa and Banco Santander.

One of the R&D areas of IIC is precisely the forecasting of wind and solar energy, where currently it services about 100 individual wind farms in Spain as well as provides global predictions over peninsular Spain to Red Eléctrica de España.

IIC`s experience in this subject ensures first an excellent opportunity to field test any potential algorithmic advances and also opens the way to exploit the most successful ones to improve current wind or solar energy predictions.