JNCASR Scientists Develop Efficient, Cost Effective Network Mimicking Human Brain
The human brain. (Wikimedia Commons photo)
By N. B. HOMBAL/IAN
BENGALURU — Scientists from Bengaluru-based Jawaharlal Nehru Centre for Advanced Scientific Research have developed a device that can mimic the human brain’s cognitive actions and is more efficient than conventional techniques in emulating artificial intelligence, thus enhancing the computational speed and power consumption efficiency.
Aiming to develop a synaptic device for neuromorphic applications with a humble fabrication method, the JNCASR team explored a material system mimicking neuronal bodies and axonal network connectivity much like the biological system. In order to realize such a structure, they found that a self-forming process was easy, scalable, and cost-effective.
The JNCASR team composed of Prof G.U. Kulkarni and B. Bharath who have developed a prototype kit developed to emulate the famous Pavlov’s dog behavior clearly demonstrates the potential of the device towards advanced neuromorphic artificial intelligence.
“This is an interesting case where chemical methods have been employed to realize a synaptic device. With this, the JNCASR has moved a step further in accomplishing advanced neuromorphic artificial intelligence,” the researchers pointed out.
According to these researchers, the human brain comprises nearly a hundred billion neurons consisting of axons and dendrites. “These neurons massively interconnect with each other via axons and dendrites, forming colossal junctions called synapses. This complex bio-neural network is believed to give rise to superior cognitive abilities,” Prof. Kulkarni said.
Using programmed electrical signals as a real-world stimulus, this hierarchical structure emulated various learning activities such as short-term memory (STM), long-term memory (LTM), potentiation, depression, associative learning, interest-based learning, supervision, etc., impression of supervision. Synaptic fatigue due to excessive learning and its self-recovery was also mimicked.
“Remarkably, all these behaviors were emulated in a single material system without the aid of external CMOS circuits,” the researchers pointed out.
One of the major highlights of this research is that software-based artificial neural networks (ANN) can be seen defeating humans in games (AlphaGo and AlphaZero) or helping handle the Covid-19 situation.
“However, the power-hungry (in megawatts) von Neumann computer architecture slows down ANNs performance due to the available serial processing while the brain does the job via parallel processing consuming just 20 W. It is estimated that the brain consumes 20 percent of the total body energy. From the calorie conversion, it amounts to 20 watts. While the conventional computing platforms consume megawatts, i.e., 10 lakh watts of energy, to mimic basic human cognition,” the JNCASR researchers said in their paper published in journal Materials Horizon.
In this paper, the researchers stated that artificial intelligence is now a part of our daily lives, starting from email filters and smart replies in communication to helping battle the Covid-19 pandemic, but it can do much more such as facilitate self-driving autonomous vehicles, augmented reality for healthcare, drug discovery, big data handling, real-time pattern/image recognition, solving real-world problems, and so on.
“These can be realized with the help of a neuromorphic device which can mimic the human brain synapse to bring about brain-inspired efficient computing ability,” the researchers argued.