The researchers of the Bar-Ilan University of Israel have introduced a new kind of ultrafast artificial intelligence algorithms. The new type of algorithms is an outcome of combined use of advanced experiments on neuronal cultures and simulations based on brain dynamics. It is believed to outperform the learning rate of the current known learning algorithms.
The researchers aim to bridge the gap between neurobiology and machine learning through the use of new learning algorithms.
Currently, most people believe that neurobiology and machine learning are two completely different and independent disciplines. However, Ido Kanter, professor and lead researcher of the Department of Physics and Gonda (Goldschmied) Multidisciplinary Brain Research Center of Bar- Ilan University, is of the opinion that the lack of reciprocal influence between the two disciplines is puzzling.
Kanter compares the bits of a modern computer to the neurons in the human brain. In an ‘Issue of Scientific Reports,’ Kanter explained that the number of neurons in a human brain is lesser than the number of bits in a personal computer. Moreover, the computational speed of the brain is slower than even the very first invented computer, and the learning rules of a brain is way more complicated and remote than the learning principles of current artificial intelligence algorithms.
The current artificial intelligence algorithms can predict an output only for a well-synchronized input, whereas the human brain does not require a well-synchronized input. This is a disadvantage of brain dynamics.
But the researchers have decided to look upon this disadvantage as an advantage. Their studies show that ultrafast learning rates are the same for a large as well as a small network. They have also found that learning can take place using asynchronous inputs also, through self-adaptation.
The findings of Kanter and his team clearly reflects that a future with zero difference between neurobiology and machine learning is not too far.