Artificial intelligence makes video games more realistic and helps your phone recognize your voice, but the power-hungry programs consume a lot of energy. Thanks to computer chips that work like the human brain, the next generation of artificial intelligence may be 1000 times more energy efficient. A new study shows that such chips can use a fraction of the energy used by ordinary chips.
Steve Furber is a computer scientist at the University of Manchester. He says that such advances could lead to huge leaps in performance in complex software.
One of the most complicated things an artificial intelligence program can do is keep bits of the pattern straight as it pieces together the whole thing. Consider how a computer can recognize an image. The well-defined edges of that image are spotted first. As it forms the final picture, it must remember the edges.
A software unit called long short-term memory is a common component of such networks. As the software determines whether the door of a car is a part of the numeral 4 or not, a vertical edge in an image needs to be retained in memory. The systems need to keep track of hundreds of elements at once.
Current networks of LSTMs are very accurate. The chips are hungry. To process bits of information, they must first retrieve individual bits of stored data, manipulate them, and then send them back to storage. Then repeat that sequence over and over again.
Intel, IBM, and other chipmakers are experimenting with an alternative chip design. If the total input exceeds a threshold, each neuron in the network will start to fire. The new chips are designed to have the hardware equivalent of a network. In conventional computers, faux neurons are defined in software and reside in the computer's separate memory chips.
Our brains only need 20 watt of power, the same as an energy efficient light bulb, because of the setup in a neuromorphic chip. To make use of this architecture, computer scientists need to change how they carry out functions.
Wolfgang Maass is a computer scientist at the University of Technology. He and his colleagues wanted to duplicate a memory storage mechanism in our brains. After a neuron in the brain fires, it typically returns to its baseline level and remains quiescent until it 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 800-273-3217 In AHP networks, a dead period that helps the network of neurons retain information, is what happens after a single firing.
The AHP neuron firing pattern was integrated into the neural network software by the group. The first challenge was to recognize a handwritten note in an image. They found that their image recognition algorithm was up to 1000 times more energy efficient than the ones on conventional chips.
The authors reported this week in Nature Machine Intelligence that the computer's setup was 16 times more efficient than a conventional computer's.
The second test was done on a series of Intel's first- generation loihi chips, which consume a lot of energy in communicating with each other. He says that the company has come out with a second-generation loihi chip, each with more neurons, which should reduce the need for chip-to-chip communication and make the software run more efficiently.
There are few commercially available neuromorphic chips. Wide-scale applications likely won't emerge quickly. An expert at the Allen Institute says that advanced artificial intelligence could help these chips gain a commercial foothold.
That could lead to novel applications, such as artificial intelligence digital assistants that could prompt someone with the name of a person in a photo, but also remind them where they met and relate stories of their past together. Mass says future neuromorphic setups may one day explore how the many firing patterns in the brain work together to produce consciousness.
All Rights Reserved © 2024
