|
|
|
|
Cronograma (Schedule) - JAICC 2009:
DÍA 1 (DAY 1) – Jueves (Thursday) 23 Sesión 1: Interfaces cerebro computadoras (ICC)
Volver a cronograma - Volver JAICC09 Sesión 2: Interfaces hombre máquina (IHM)
Volver a cronograma - Volver JAICC09
DÍA 2 (DAY 2) – Viernes (Friday) 24 Sesión 3: Interfaces cerebro computadoras (ICC)
Volver a cronograma - Volver JAICC09 Resúmenes Conferencias- JAICC 2009:
Prof. Dr. José del R. Millán
Day 1: An Overview of BCI: From Control to Monitoring: The idea of moving robots or prosthetic devices not by manual control, but by mere "thinking" (i.e., the brain activity of human subjects) has fascinated researchers for the last 30 years, but it is only now that first experiments have shown the possibility to do so. Such a kind of brain-computer interface (BCI) is a natural way to augment human capabilities by providing a new interaction link with the outside world and is particularly relevant as an aid for physically disabled people. In this talk I'll give an overview of the field of non-invasive BCI, which can exploit either evoked potentials or spontaneous brain phenomena recorded with surface electrodes placed on the subject's scalp. Key elements for a successful BCI are real-time feedback and training, of both the subject and the classifier embedded into the BCI. Normally, a people use a BCI to control and interact with devices by constantly delivering mental commands. But a BCI can also provide an effective way to monitor some cognitive states of the subject that, if recognized in real time, can improve and facilitate tremendously interaction.
Day 2: A Look Behind the Veil of an Adaptive Brain Interface: In this talk I will review our work on non-invasive asynchronous BCI, with a focus on how brainwaves can be used to directly control robots. Most of the hope for such a possibility comes from invasive approaches that provide detailed single neuron activity; however, it requires surgical implantation of microelectrodes in the brain. For humans, non-invasive systems based on electroencephalogram (EEG) signals are preferable but, until now, have been considered too poor and slow for controlling rapid and complex sequences of movements. Recently we have shown for the first time that online analysis of a few EEG channels, if used in combination with advanced robotics and machine learning techniques, is sufficient for humans to continuously control a mobile robot and a wheelchair. Finally, we discuss current research directions we are pursuing in order to improve the performance and robustness of our BCI system, especially for real-time control of brain-actuated robots. In particular, I'll mention work on recognizing cognitive states that are crucial for interaction. Volver a cronograma - Volver JAICC09
Prof. Dr. Enrique Spinelli LEICI- Dto de Electrotecnia Universidad Nacional de La Plata y CONICET - Argentina Plataformas para Implementación de Interfaces Cerebro-Computadora (ICC) en Tiempo Real: Las ICCs se instalaron como tema de investigación del LEICI alrededor del año 1997 y en 2000 se ensayó la primera ICC completa. Este primer equipo comprendía el amplificador de EEG, un adquisidor de señales, procesamiento digital de señales en tiempo real y actuadores para manejar electrodomésticos como luces y ventiladores, los cuales eran activados a partir de ritmos cerebrales. De aquí en más, las investigaciones se centraron en el desarrollo de plataformas para implementación de ICC en tiempo real. A lo largo de la charla se describirán los trabajos realizados en los últimos 10 años con el objetivo de implementar plataformas autónomas para ICCs robustas y portables. Los desarrollos efectuados, intentando siempre perseguir el “estado del arte” y proponer soluciones originales, abarcan diversas partes de las ICC como el amplificador de EEG, la fuente de alimentación, la transmisión de datos, el software de tiempo real y el de presentación. La presentación intenta, en un desarrollo cronológico, mostrar la evolución de las técnicas y de las tendencias a lo largo del tiempo; presentando la experiencia propia con sus resultados satisfactorios y aquellos capitalizados como aprendizaje (mal llamados fallidos). Por ejemplo, se describirá el desarrollo del amplificador de biopotenciales, el uso de convertidores Sigma-Delta de alta resolución y su evolución desde un circuito clásico alimentado con ±15V y con un gran número de componentes, a amplificadores de un reducido número de elementos, alimentados con una fuente simple de 3V. Se mostrará también la evolución del software y de la transmisión de datos desde una primera solución opto-acoplada a la transmisión inalámbrica mediante tecnologías Bluetooth y Zigbee, pasando por la utilización de fibras ópticas. Finalmente se discutirá la investigación en ICC como un interesante eje para el desarrollo de recursos humanos en procesamiento de señales y desarrollo de hardware, porque al mismo tiempo que requiere imperativamente el desarrollo de estas técnicas, proporciona un objetivo claro, fascinante y motivador para la aplicación de las mismas.
Volver a cronograma - Volver JAICC09
Dr. Laurent Bougrain INRIA project-team CORTEX Nancy University - France Template-based classifiers for ERP-based BCIs: This talk aim to present pattern recognition techniques of graphic elements (e.g. event-related potential, auditory evoked potential, k-complex, sleep spindles, vertex waves) included in electro-encephalographic signals. More specifically, template-based classifiers will be introduced to robustly detect evoked potentials in a single trial from noisy and multi-sources electro-encephalographic signals. Brain-computer interface (BCI) system is a potentially powerful new communication and control option for those with severe motor disabilities [Wolpaw et al., 2002]. A BCI system translates brain activity into commands for a computer or other devices (e.g. wheelchair, robotic arm). In other words, a BCI allows users to act on their real or virtual environment by using only brain activity. One of the well-known and powerful BCI system is the P300 speller based on the non-invasive Electroencephalography (EEG) measuring from the subject's scalp [Farwell & Donchin, 1988]. This BCI system uses an oddball paradigm. Oddball paradigms are used in asynchronous BCIs to induce event-related potentials (ERPs) through visual or auditory stimulus. Subject pays attestation to specific stimuli that will induce ERPs when presented; others are regarded as the background neural activities. One well-defined component of ERPs is the P300 component that is a positive deflection waveform observed around 300ms after the onset of the stimulus. Brain-Computer interfaces based on evoked potentials allow more commands than the ones based on mental tasks and do not need long human training. Almost everybody reacts on them and there are used by patients. Thus, the task of the P300 speller system is to recognize the ERP components from the noisy EEG background signal. It is found difficult to accomplish this target on the base of a single trial because the magnitude of the EEG background activities is usually one-order larger than the one of the ERP components, that means the ERP components in single-trial recordings are almost covered by the background neural activities. Moreover, non-invasive electrodes produce a noisy signal because the skull dampens signals. Thus, ERP detection usually needs to average responses of repeated stimulations. Due to the averaging operation, the background EEG activities are reduced and the ERP components are enhanced and evident. From a practical point of view, an important issue is to reduce the number of repetitions, in order to obtain high communication bitrates. The methodology has been improved but a gap still exists to enable single-trial recognition. Template-based methods with alignment techniques look potentially interesting to obtain a robust detection. ERPs are short-time events with characteristic peaks at specific times. So it is useful to be able to extract features in time domain. Template-based classifiers first estimate ERP templates by averaging in the time domain ERP responses, and then use the shorter distance between the current response and the ERP templates as the discriminant criterion. We will focus our attention in particular on several averaging techniques and distance measures such as the point-to-point averaging, the cross-correlation distance [Wastell, 1977] and the dynamic time warping [Kang et al., 2001]. One of these methods, which evokes our attention to solve this problem, is the learning vector quantization algorithm (LVQ) [Kohonen, 1990]. LVQ is able to automatically extract morphology-specific templates. It performs a supervised learning for a classification task. Its principle is first to cluster input data using a competitive learning without any spatial relationship between codebook vectors (i.e., templates). Each cluster is pre-assigned to a specific class. Thus, when a new pattern should be classified, the method determines in which cluster the pattern belongs to and then assigns it to the corresponding class label. This machine learning technique is well known for its robustness and interpretability. However, for large-scale complex problems, its performance can be further improved by optimizing the way to combine the templates in its assignment stage instead of selecting one of the templates as the final decision. More precisely, it could be interesting to overcome the simple identity function to determine classes from clusters. Thus, a non-identity LVQ will be described. This improvement will be accomplished using the same scheme found in the extreme learning machine algorithm (ELM) [Huang et al, 2004]. The experimental results show that the proposed algorithm improves the accuracy with less computational units compared to original LVQ and ELM.
[Farwell and Donchin, 1988] L.A. Farwell and E. Donchin. Talking off the top of your head: toward a mental prosthesis utilizing event-related brain potentials. Electroencephalography and Clinical Neurophysiology, 70(6):510–523, 1988. [Huang et al, 2004] G.-B. Huang, Q.-Y. Zhu, and C.-K. Siew. Extreme learning machine: A new learning scheme of feedforward neural networks. In Proceedings of International Joint Conference on Neural Networks, volume 2, pages 985–990, Budapest, Hungary, 2004. [Kang et al., 2001] K. Wang, H. Begleiter, and B. Porjesz. Warp-averaging event-related potentials. Clinical Neurophysiology, 112(10):1917–1924, 2001. [Kohonen, 1990] T. Kohonen. The self-organizing map. In Proceedings of the IEEE, volume 78, pages 1464–1480, 1990. [Wastell, 1977] D.G. Wastell. Statistical detection of individual evoked responses: an evaluation of woody’s adaptative filter. Electroencephalography and Clinical Neurophysiology, 42:835–839, 1977. [Wolpaw et al., 2002] J. R. Wolpaw, N. Birbaumer, D. J. McFarland, G. Pfurscheller, T. M. Vaughan. Brain–computer interfaces for communication and control. Clinical Neurophysiology, 113:767–91, 2002.
Volver a cronograma - Volver JAICC09
Mg. Ing. Rodrigo Salas F. Departamento de Ingeniería Biomédica Universidad de Valparaíso - Chile Robust and Incremental Neuro-fuzzy Techniques for Evoked Potential Detection in BCI: Brain-computer interface (BCI) systems are new options for communication that operate through the translation of brain activity to commands in the computer or other devices. In other words, BCI users are allowed to operate in real or virtual environments using only brain activity. The evoked potential of electroencephalographic (EEG) signals is one of the most studied non-invasive interfaces, mainly due to its fine temporal resolution, easy to use and low cost of use. Unfortunately, non-invasive implants produce noisy signals because the signal to noise rate reduction by the skull. Another major hurdle in the use of EEG for brain-computer interface is the excessive training time required before the user can use the technology. In this talk, we will present various computational intelligence techniques that can be used for robust and efficient detection of the “activated” regions of the cortex from electro-encephalographic noisy signals obtained from multiple sources. Adaptive techniques based on artificial neural networks and neuro-fuzzy systems will be presented. In particular, neuro-fuzzy models are characterized by their ability to model systems with the flexibility and interpretability of the fuzzy inference systems, and adaptation of neural networks. We will explain theoretical aspects to designing robust learning algorithms to estimate event-related potentials (ERP) from highly noisy EEG signals and where the behavior may be different for each user. We will present incremental learning techniques to enable neuro-fuzzy models to automatically analyze large amount of data that arrive sequentially in time in form of batch, and, where a balance between the previous and currently learned patterns will be preserved. The construction of incremental learning algorithms do not require a priori definition of the architecture and the models will search for the ad-hoc architecture to model as best as possible the underlying structure of the data.
Volver a cronograma - Volver JAICC09
Ms.C. Bioing. Gerardo Gabriel Gentiletti
La propia definición del nombre adoptado para
identificar esta apasionante línea de investigación: Interfaces Cerebro
Computadora -ICC- (Brain Computer Interface -BCI- en inglés) nos sugiere cual
fue el principal eje de investigación sobre la misma, el cuál ha sido el de
lograr establecer "un canal de comunicación entre el cerebro (la persona misma)
y una computadora", sin requerir de otra vía natural (músculos o actividad en
nervios periféricos). Sin embargo, a la par del avance tecnológico, las
posibilidades de comunicación se han extendido en los últimos años a
posibilidades de accionar o interactuar físicamente con el medio que rodea a la
persona que se conecta a una ICC. La forma de extender de manera natural las
capacidades de la computadora misma para lograr este accionar, actualmente está
en manos principalmente de la Robótica. Sin embargo a la propia complejidad que
implica el sólo hecho de poner a funcionar una ICC, debemos agregar ahora la
complejidad de los propios dispositivos robóticos que se pretenden "comandar",
donde sillas de ruedas inteligentes, brazos manipuladores, móviles
telecomandados y actuadores de domótica, son algunos de los ejemplos que podemos
citar.
Volver a cronograma - Volver JAICC09 | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|