Intel wants to revolutionize artificial intelligence systems with the first neuromorphic self-learning microprocessor.
Artificial intelligence continues to advance small cases and facilitate tasks, although it still does not reach very high levels of development as science fiction fanatics would like to see, there are very useful case examples as in the field of medicine. So every time that an improved system appears is good news, this time from the hand of the company Intel.
The technology has announced the launch of the first neuromorphic self-learning chip in its category. To date, there has been much talk about machine learning and the training of a machine to interpret a lot of data that humans would not be able to do with that speed, but not the word “neuromorph” in a microprocessor. The company promises that this chip mimics the human brain.
“Loihi” – the name of this chip – learns through the interpretation of the different types of response to the stimuli of an environment. Like other systems, it uses data to learn, but with the difference that it does not need to be trained as it normally does, thanks to a novelty in its development.
“We believe that AI is in its initial phase and that more architectures and methods, like the ‘Loihi’ chip, are going to continue to raise the AI level. Neuromorphic computing draws on current knowledge of brain architecture and its associated computations, “said Michael Mayberry, corporate vice president and general manager, Intel Labs.
The autodidact training that is explained with another of its great novelties, while the machines follow guidelines like “1 + 1 = 2” and they do not know to operate before something for which it has not been programmed to them, the system puts changes or errors to him to which by itself learns to discern it, very similar to the neural networks of the human brain. Technically, this process is called as asynchronous spike activity potentials (spikes).
«Las redes neuronales del cerebro obtienen información mediante impulsos eléctricos o potenciales de acción de actividad eléctrica, y modulan las fuerzas sinápticas o el peso de las interconexiones basándose en la duración de estos potenciales de acción, guardando estos cambios de forma local en las interconexiones. Las conductas inteligentes surgen de las interacciones cooperativas y competitivas entre múltiples zonas dentro de las redes neuronales del cerebro y de su entorno», ha indicado Mayberry.
The microprocessor has a neuromorph network of 130,000 neurons, has several cores each with a learning engine and is thought to be more efficient. Learning models, such as “deep learning”, which do not initially have circumstances or bland elements, do not know how to generate an answer because they do not know how to function in disordered environments. So systems that know distinguish abnormal patterns may have different implications as in medicine to notice atypical latinos in the heart.