Centre for Sensors, Instruments and Systems Development

Universitat Politècnica de Catalunya

Shaping light to your needs

Foto Foto Foto Foto



Seminar: Machine Learning for data analysis and visualization

Under the project DAVALOR, CD6 organizes a seminar titled "Machine Learning for data analysis and visualization" taught by Dr. Jesus Malo, professor at the University of Valencia. More information about the speaker and his group can be found at http://isp.uv.es.

The seminar will take place on November 21st at 12,30h in the Joan Salvadó Auditorium (Passeig 18 de Juliol 660, Terrassa)


Machine learning is the new name given to the statistical techniques (some new and some not so) for extracting information from data. This "new" discipline (based on the raw data knowledge rather than theorizing) has attracted interest in the last decade for two reasons: (1) as general point of view, and (2) another linked to the learning concept itself.

From a general point of view the interest is not so much the superiority of empirical knowledge versus theoretical knowledge, but the massive accessibility to data currently available (big data), as well as the increase in computing power (big CPU). On the other hand, in the case of fostering to understand the phenomenon of learning, it goes beyond the (naive) big-data / big-CPU. In the latter case, the question (cognitive problem) is to understand how a system like the brain can develop an abstract knowledge from specific perceptions.

Besides giving examples of this definition, in this talk the overall Hastie et al. program will be discussed  (Hastie, Tibshirani, Freeman. The Elements of Statistical Learning. 2nd Ed. Springer Verlag. Berlin. (2009) http://statweb.stanford.edu/~tibs/ElemStatLearn/, with emphasis on different techniques including examples on NO SUPERVISED LEARNING (feature extraction based on statistical independence and efficient coding, as Principal Components Analysis or Independent Components Analysis, Kernel-PCA, dimensionality reduction, clustering, vector quantization and Self Organizing Maps, transform coding, principal curves, probability density estimation (Projection Pursuit, Gaussianization), measures of statistical dependence (Mutual Information and HSIC), and examples of SURPERVISED LEARNING in regression and classification (Support Vector Machines, Kernel Ridge regression, LASSO -Least Absolute Shrinkage and Selection Operator-, Model selection by Cross -validation, Total Variation Regularization, Neural Networks and Autoencoders).
CD6 Centre for Sensors, Instruments and Systems Development
Rambla de Sant Nebridi, 10  ·  08222  ·  Terrassa (Barcelona)