Neural networks learn faster thanks to IBM technology
Artificial intelligence is a useful but resource-consuming technology. In an attempt to speed up the neural network training with new knowledge, specialists often go beyond its optimal energy consumption. As a result, working with AI can simply become unprofitable.
New IBM technology is designed to solve this problem. At NeurIPS 2018 conference held in early December, the American company presented several developments that can reduce the level of energy consumption when working with neural networks.
The problem of energy consumption by artificial intelligence
Now 32-bit and 16-bit computing technology is used for the development of neural networks. It guarantees a high speed of AI training, but there is also a reverse side of the coin — the higher the speed, the more energy goes away. The high level of energy consumption in the context of the development of artificial intelligence make the work of specialists extremely expensive.
IBM technology: what, how and why
Employees of IBM corporation decided to abandon the fast, but not cheap technology, and created a more cost-optimized technique.
The company's specialists have completely changed the hardware components of the neural network. Within IBM Research program, they also changed the format of its work. Now AI is not used to solve specific problems, and is able to act broadly and affect different areas.
At NeurIPS 2018, IBM researchers presented two developments. The first one declared that the company managed to maintain a 16-bit level of accuracy on 8-bit system. At the same time, the speed of AI learning is several times higher than that of 16-bit.
The second development is the most important. As it reveals, due to what IBM was able to reduce the level of energy consumption by 33 times. Low accuracy analog chips is compensated for high performance training in the niches with the use of energy.
Apparently, IBM technologies in the future will make the use of neural networks more affordable and cost-effective from the financial point of view. This means that not far off another news about a new milestone in the development of artificial intelligence.