IBM created a computer chip that seeks to mimic the functioning of synapses in the brain, which paves the way for a machine capable of perceiving, learning, and in some situations, act like humans.
In IBM's view, computers with this type of chip - the company calls cognitive computers - "will not be programmed the same way that traditional computers." Instead, "is expected to learn from the experiences, find correlations, create hypotheses and memorize - and learn from - the results," said IBM in a statement.
In the brain, the synapses connecting the neurons to other cells including other neurons. These links can be strengthened, weakened or eliminated, allowing the brain to learn and adapt. Synapses are essential in the process of perception and thought.
It is this flexibility of connections that the IBM scientists are seeking to emulate the new chips.
The company did not elaborate on the technology works, but explains that "through advanced algorithms and circuits," chips are able to weaken or strengthen the links between its components, according to the information they receive.
Currently, there are systems capable of learning, but this process takes place at the software level rather than at the underlying hardware.
The system developed by IBM consists of two chips, each with 256 "electronic neurons" that are arranged in parallel and connected by "synapses" one of these chips has 262,144 "synapses" that are programmable and the other has 65,536, capable of learning. Although the circuit is not able to physically change the connections between components, can evolve to give priority to some links over others.
Unlike in conventional computers, these "neurons" (which constitute the system processor) are very close to the "synapses" (which is memory), which allows each connection consumes much less energy than usual - a the characteristics of the brain is precisely the reduced energy consumption given the enormous amount of links that do.
"The long term goal," IBM says, "is to build a system chip with ten billion neurons and 100 trillion synapses, consuming only one kilowatt of power and occupying less than two liters of volume." Estimates for the number of neurons in the human brain vary. Some point to values of around 80 billion neurons, with other counts to suggest a number above 100 billion.
The company is already testing prototypes and design, which includes contributions from several American universities, received funding from DARPA, a research agency in dependence on the U.S. Department of Defense.
In IBM's view, computers with this type of chip - the company calls cognitive computers - "will not be programmed the same way that traditional computers." Instead, "is expected to learn from the experiences, find correlations, create hypotheses and memorize - and learn from - the results," said IBM in a statement.
In the brain, the synapses connecting the neurons to other cells including other neurons. These links can be strengthened, weakened or eliminated, allowing the brain to learn and adapt. Synapses are essential in the process of perception and thought.
It is this flexibility of connections that the IBM scientists are seeking to emulate the new chips.
The company did not elaborate on the technology works, but explains that "through advanced algorithms and circuits," chips are able to weaken or strengthen the links between its components, according to the information they receive.
The system developed by IBM consists of two chips, each with 256 "electronic neurons" that are arranged in parallel and connected by "synapses" one of these chips has 262,144 "synapses" that are programmable and the other has 65,536, capable of learning. Although the circuit is not able to physically change the connections between components, can evolve to give priority to some links over others.
Unlike in conventional computers, these "neurons" (which constitute the system processor) are very close to the "synapses" (which is memory), which allows each connection consumes much less energy than usual - a the characteristics of the brain is precisely the reduced energy consumption given the enormous amount of links that do.
"The long term goal," IBM says, "is to build a system chip with ten billion neurons and 100 trillion synapses, consuming only one kilowatt of power and occupying less than two liters of volume." Estimates for the number of neurons in the human brain vary. Some point to values of around 80 billion neurons, with other counts to suggest a number above 100 billion.
The company is already testing prototypes and design, which includes contributions from several American universities, received funding from DARPA, a research agency in dependence on the U.S. Department of Defense.
The technology could be applied in situations where a computer has to collect a large amount of unstructured information (which may include visual information, sound or temperature, for example), process it immediately, as humans do, and choose a given action.
One possible application would be the management of road traffic. Computers are already good at analyzing large amounts of data, since these are the right format. But they have difficulty perceiving an environment filled with different information, scattered and unpredictable, like the sound of children running, the image of the car floor, lights traffic lights and horns.
Source publico.pt
Comments :
0 comments to “IBM develops chip inspired by the functioning of the brain”
Post a Comment