Sunday, July 25, 2010

SMART COMPUTERS



Scientific outpost in computation is copying in parallel animal biologic networks in order to create computers able to see, to hear, to make decisions and until being emotion moved. Advance and backwardness, because to copy is not to innovate. Copies are not similar to the original. Real innovation would be to design a computer (quantum or other), able to compute instantly the origin, life and death of a storm, to know instantly the final form and product of a vegetable starting from the computation of its genome or the structure of a unique leaf; to establish the whole structure of our unique satellite, with the simple computation of the sway of earth marine waves. Thomas Wennekers (U. Plymouth), tries to build computers that see and listen learning how neurons communicate among them. For it he simulates a groups of neurons and laminar microcircuits to great scale of visual neocortex (implied in seeing and listening), crucial fact to recognize objects and words.

For their side Steve Furber (U. of Manchester), has created the software SpiNNaker with multiple chips that simulate a great number of neurons and whose only communication with the rest of the world is a little signal or sound that send information to a net of small computers. Each arm processor of SpiNNaker, simulates 1,000 neuronal models. Furber tries to design a chip with 18 arms processors, 16 of which will models neurons that will learn to see. The goal of Furber is to create a system composed of a billion neurons, controlled by a million of processors ARM. The topic is to understand how the brain passes a very big lot of computational elements from a small area to another, managing the system without being surpassed by superior coordination of these processors. A way to do that is that individual processors use less power, be every time but small and be endowed with self-organization capacity.

COMPUTADORAS INTELIGENTES

La avanzada cientifica en computación esta copiando las redes biologicas neuronales en paralelo de animales a fin de crear computadoras capaces de ver, oir, tomar decisiones y hasta emocionarse. Avance y atraso, porque copiar no es innovar. Las copias no son iguales al original. Creación seria diseñar una computadora (cuántica u otra), capaz de computar instantáneamente el origen, vida y muerte de una tormenta en curso o el producto final de un vegetal a partir de la computación de la estructura de una hoja; establecer la estructura total de nuestro único satélite, con la simple computación del vaivén de las olas marinas terráqueas. Thomas Wennekers (U. Plymouth), intenta construir computadoras que vean y escuchen aprendiendo como se comunican las neuronas entre si. Para ello simula grupos de neuronas y microcircuitos laminares a gran escala del neocortex visual implicados en ver y escuhar, hecho crucial para reconocer objetos y palabras.

Por su lado Steve Furber (U de Manchester), ha creado el software Spinnaker con multiples chips, que simulan gran numero de neuronas y cuya unica comunicación con el resto del mundo son señales o sonidos que envían información a una pequeña red de pequeñas computadoras. Cada brazo procesador del Spinnaker, simula 1,000 modelos neuronales. Furber intenta diseñar un chip con 18 brazos procesadores, 16 de los cuales modelaran neuronas que aprenderan a ver. La meta de Furber es crear un sistema de un billón de neuronas, controladas por un millón de procesadores ARM. El tema es comprender como el cerebro pasa un lote muy grande de elementos computacionales de una pequeña area a otra, manejando el sistema sin ser sobrepasado por la coordinacion superior de estos procesadores. Una via es que los procesadores individuales usen menos poder, sean cada vez mas pequeños y sean capaces de autoorganizarse.

No comments: