The neural networks based on IBM processor grows, and has 48 million digital neurons
A year ago, IBM announced the creation of a different processor than we had known. According to them, they had managed to produce a design inspired by the Brain and working as a neural network chip ( here we explain everything you need to know about them ) rather than as a conventional computer. This week announced a step forward on this project.
IBM explains that have managed to create a system with 48 million digital neurons. This figure approaches to having a small rodent. In the US company it is clear that the future of computing is to create two types of computers those that work conventionally and work with those synaptic connections.
Dharmendra Modha, head of research at IBM, he wants this new chip is an evolution in what is known in the computer world. The reality is that neural networks and their use is not exactly new. According to him, are devices that can do calculations more quickly, if not with specific data or a comparison has been made yet.
Instead of analyzing the information in sequential order, they want this processor acts as if the right brain and is able to find patterns in data to streamline tasks and be more efficient.
The configuration is different from a conventional chip. Instead of executing instructions, the chip handles a series of peaks (spikes) with simple information packages that are similar to electrical pulses in the brain. In each peak can be displayed, for example, changes in a person’s voice when speaking or changes in color of a pixel in a photograph to another. If you want further explanation, while it will detail in this article.
The programming and work with these chips is, the company says, very different from today, today is carrying out campaigns with research centers, and government agencies to explain how they work and how to be applications do. For them, the way we are working now is like trying to sneak a round card by a square hole.
This project is not new and IBM have been working many years on it. In 2011 they got their capacity out as the brain of a bee . Four years after that event, now it has almost as many neurons as a mouse, although being digital is difficult to make a comparison between them. Their goal? To create a network of 10 billion neurons in digital “brain” to enter into a shoebox.
Although this project IBM is fascinating, it can happen over time to become the new Watson. One promising technology that fails to boot. Today raises several questions: no applications or forms of contrast that this model improves the von Neumann architecture. Although they sell neural networks as something new, we have been working with them for many years. In fact, IBM’s own research with them from the 50. Without evidence, which have been put on the table does not invite to think that will change the landscape.