Researchers from Centrum Wiskunde & Informatica (CWI) has made a mathematical breakthrough. They have developed a learning algorithm for the spiking neural networks inspired by the human brain. It is said that the discovery can make AI algorithms a thousand times more energy-efficient. And to efficiently operate spiking a neural network, a new type of chip is required. It is said that different companies are trying to develop prototypes.
Thanks to the team of researchers from Centrum Wiskunde & Informatica (CWI) who have made a mathematical breakthrough, AI apps like speech and gesture recognition and ECG classification have the potential to become a thousand times more energy-efficient.
The current AI apps that are out there aren’t energy-efficient enough to process data on smart devices locally. Thus, you see speech recognition and gesture recognition apps relying on cloud connection to operate.
But now, because of the breakthrough, you can put much more elaborate AI in chips. It will enable apps to run on smart devices like phones and watches. You will not only have more robust apps but have more privacy as data can be stored and processed locally.
Researchers of Centrum Wiskunde & Informatica (CWI), the Dutch national research center for mathematics and computer science, made the mathematical breakthrough while working collaboratively with the IMEC/Holst Research Center from Eindhoven, The Netherlands.
You will find the research findings published in a paper by Bojian Yin, Federico Corradi, and Sander M. Bohté. They have also made the underlying mathematical algorithms open source. So, if you are interested, you can easily avail of it.
Let’s find out about the revolutionary algorithm, what inspired Bohté, and the novel computer chip needed to run spiking neural networks.
Revolutionary new algorithm
It was under the supervision of Sander Bohté, researcher of CWI and UvA professor cognitive neurobiology, that researchers developed a learning algorithm for the supposed spiking neural networks.
Spiking neural networks are not new, but they are challenging to manage from a mathematical perspective. Thus, putting it into practice is difficult.
The new algorithm is revolutionary, as there is no need for the neurons in the network to communicate a lot. Plus, each individual neuron has to perform less calculation.
It is the combination of these two discoveries that make AI algorithms a thousand times more energy-efficient.
Inspired by the brain of the human
It is said that Bohté got his inspiration from the unbelievably energy-efficient way in which information is possessed by the human brain.
Computers mimicking the brain’s neural networks have created incredible apps such as speech recognition, image recognition, medical diagnoses, etc., in recent years, but they needed up to a million times more energy as compared to the human brain.
The new spiking neural networks are different in that they are more similar to the human brain. They also communicate only less and with short pulses. It also means that the signals are sporadic and not easy to handle mathematically.
Novel computer chip to run spiking neural networks
In order to operate spiking neural networks efficiently, there is a need for a new type of chip. Bohté mentioned that different companies are in the process of developing prototypes. They are all in it together.
The methods applied by Bohté can train spiking neural networks encompassing a few thousand neurons (fewer than usual neural networks), but enough for several apps like ECG classification, speech recognition, and gesture recognition.
They also plan to scale up these networks to 100.000 neurons to expand the app possibilities further.
Aside from its contribution to the neural network, programming, and other similar fields, energy efficient AI also helps you improve your SEO, build optimized apps and sites, identify your audience, and more.
This article has been published from a wire agency feed without modifications to the text. Only the headline has been changed.
