jeudi 8 août 2013

IBM wants you to learn how to program its brain chips



IBM Research has created programming software for its experimental brain chips.

Big Blue’s researchers in San Jose, Calif., have created what they call a “breakthrough software ecosystem” designed for programming chips whose architecture has been inspired by the way the brain functions. The chips function on low power and are good at applications that the brain can handle, like perception, action, and cognition.

The software is dramatically different from traditional software. IBM says its new programming model breaks the model of sequential operation that traces its roots back to the John von Neumann architecture created in the 1940s. Instead of handling one instruction at a time, IBM’s new model is tailored for doing lots of things at once. The new model is made for distributed, highly interconnected, asynchronous, parallel, and large-scale cognitive computing architectures.

“Architectures and programs are closely intertwined and a new architecture necessitates a new programming paradigm,” said Dharmendra Modha, principal Investigator and senior manager at IBM Research. “We are working to create a FORTRAN for cognitive computing chips. While complementing today’s computers, this will bring forth a fundamentally new technological capability in terms of programming and applying emerging cognitive systems.”

IBM announced its new brain-like computer chips in 2011. But to make use of them, it has to create software that programmers will use to program the new chips. IBM believes that these brain-like computers will be better at tasks such as vision and perception, at least in comparison to “number cruncher” von Neumann computers that are the foundation of all of today’s major computing architectures, such as Intel’s x86 architecture. IBM says that the brain-like computers can get a lot more work done simultaneously, without using too much power. Working on a project funded by the Defense Advanced Research Projects Agency (DARPA), IBM demonstrated working chips based on the new technology. More than $53 million has been poured into it so far.

IBM researchers developed a number of breakthroughs aimed at making brain-like computing easier. They created a simulator to demonstrate multi-threaded, massively parallel, and highly scalable software running on a network of brain-like cores known as Synapse processors. IBM’s long-term goal is to build a chip system with ten billion neurons and hundred trillion synapses, while consuming merely one kilowatt of power and occupying less than two liters of volume.

The software follows a neuron-like model. It is a simple, digital, highly parameterized spiking neuron model. A network of such neurons can sense, remember, and act upon a variety of programming inputs. The programming model simplifies this complex structure, creating a high-level description that is based on composable and reusable building blocks called “corelets.” Each corelet is a blueprint of a network of neurons. The inner workings of the neuron are hidden from the programmer, who only sees its external inputs and outputs. So the programmer can focus on what the neuron does, rather than how it does it. Corelets are hierarchical. They can be used to create new corelets that are more complex.

The researchers have created more than 150 corelets already as part of a library that programmers can tap. IBM is presenting the programming model at the International Joint Conference on Neural Networks.



By: Dean Takahashi

Aucun commentaire:

Enregistrer un commentaire