Keynote speakers

Nicolás García-Pedrajas

Katherine Cameron,

Marc Van Hulle

Chao He

Michel Verleysen

 

Nicolás García-Pedrajas

Home Page

Spain npedrajas@uco.es

Universidad de Córdoba

 

Scaling up instance selection algorithms by means of democratization

In the last few years the problems faced by data mining algorithms have been growing in size from a few thousands of samples to hundreds of thousands or millions. This has rendered many algorithms unusable due to their scaling problems. Scaling up data mining algorithms has become an important issue, because for those algorithms "size does matter"... This talk is devoted to the methods for scaling up instance selection algorithms. Firstly, we report the three general approaches used so far for scaling up data mining algorithms: designing fast algorithms, partitioning the datasets and using relational representations. We present some of the most common methods based on these three approaches. Secondly, we present a new paradigm, the democratization of algorithms, which is able to scale up learning algorithms without any modification. It is based on a divide-and-conquer strategy coupled with the principle of the ensembles of classifiers consisting of combining weak solutions to obtain a strong one. This paradigm has been successfully applied to instance selection and feature selection with notable results in terms of reduction of the execution time of the algorithms. Finally, we present the conclusions of the talk and the open research lines in the field of scaling up data mining algorithms. [BackToTop]

Katherine Cameron

Home page

k.cameron@ed.ac.uk

The University of Edinburgh

[BackToTop]

Chao He

ch185031@ncr.com

NCR Labs

 

Pattern Recognition Applications in Self-Service

Recent global research confirms that today’s time conscious consumers are demanding more self-service options. They expect a fast, easy, personalised and convenient interaction experience with businesses. Built upon its 125 year experience in serving customers, NCR is a global leader in the self-service industry. NCR Labs, NCR's centralised research arm, drive technology innovation and applied research initiatives with a view to addressing major industry issues in self-service. Among all their research areas, pattern recognition and image analysis technologies play an important role in solving a variety of self-service challenges such as automated document analysis, object recognition, fault prediction and others across the financial, retail, travel, and entertainment industries. This talk starts with an overview of NCR Labs latest research in computational intelligence; then continues further to focus on one of most interesting and difficult challenges in financial industry by providing a technology review on automated currency validation. [BackToTop]

Michel Verleysen

Home Page

esann@uclouvain.be

Université catholique de Louvain

[BackToTop]

Marc Van Hulle

Home Page

Laboratorium voor Neuro- en Psychofysiologie
K.U.Leuven
BE-3000 Leuven
Belgium

marc_AT_neuro.kuleuven.be Katholieke Universiteit Leuven

 

Invasive Brain-Machine Interfaces in monkey visual cortex

Brain-machine interfaces (BMI's) (or brain-computer interfaces (BCI's), neuroprosthetic devices, neuromotor- or mental prostheses) have witnessed an enormous evolution in recent years. They are now generally regarded as one of the most promising engineering applications in the neurosciences. Indeed, such devices offer an immediate perspective to patients suffering from serious neurological diseases such as CerebroVasculair Accident CVA) (brain hemorrhage, stroke), Amyotropic Lateral Sclerosis (ALS), traumatic brain lesions,..., to improve their quality of life. The invasive BMI's have electrodes, or even arrays of hundreds of microelectrodes implanted mostly in the motor- and premotor cortical areas of the brain. A more recent development are the BMI's implanted in the visual cortex. In this talk, we will consider the decoding of invasively recorded brain activity from monkey visual area V4. The recordings were done using an array of 100 microelectrodes, covering an area sized 4 by 4 mm. The decoding of action potentials (a transient change in the electrical potential of a neuron's membrane) and local field potentials (a composite extracellular potential originating from hundreds to thousands of neurons around the electrode tip) will be discussed. We have used feature selection as well as feature extraction techniques, and applied linear- as well as non-linear classifiers, and binary- and multiclass classifiers. We have developed new types of secondary features for local field potentials between electrodes, based on phase-based synchrony and the speed and direction of wave propagation over the array, and showed their contribution in improving the decoding performance over the first-order features (based on single electrodes). Finally, we will discuss the putative connection between the observed increase in synchrony and perceptual learning and attention. [BackToTop]