A brain-inspired AI technology boosts efficiency and reduces energy consumption

Researchers at FORTH have developed a brand new sort of synthetic neural community (ANN) that includes options of organic dendrites. This revolutionary design permits for correct and sturdy picture recognition whereas utilizing considerably fewer parameters, paving the way in which for extra compact and energy-efficient AI programs.

Synthetic Intelligence (AI) performs an important position in driving innovation and enhancing effectivity throughout numerous industries, providing smarter options to complicated issues and enhancing our every day lives. Nonetheless, present AI programs are big, comprising millions-to-billions of parameters, thus consuming large quantities of power, which limits their widespread use.

By integrating neuro-inspired options into AI, we will create smaller and smarter programs that mimic how our brains course of info, enhancing their effectiveness in recognizing patterns and making selections. This results in extra environment friendly and efficient AI functions.

Dendrites are the branched extensions of nerve cells that resemble tree branches. Their important operate is to obtain info from different neurons and transmit it to the cell physique. For a few years, the position of dendrites in info processing was unclear, however current research have revealed that they’ll carry out complicated calculations independently of the primary neuron. Moreover, dendrites are important for the mind’s plasticity, which is its means to adapt to altering environments.

In a current article published within the journal Nature CommunicationsDr. Panayiota Poirazi’s crew on the Institute of Molecular Biology and Biotechnology (IMBB) of FORTH proposed a novel structure for synthetic neurons that includes completely different options of organic dendrites, and examined it in numerous picture recognition situations.

The findings present that these dendritic ANNs are extra proof against overfitting and might match or exceed the efficiency of conventional ANNs whereas utilizing a lot fewer sources, specifically trainable parameters and studying steps.

This enchancment arises from a singular studying strategy, whereby a number of nodes within the community contribute to the encoding of various classes. That is opposite to conventional ANNs, whereby most nodes are category-specific. Total, the analysis means that incorporating dendritic options could make ANNs smarter and extra environment friendly.

This work was led by Dr. Chavlis, a postdoctoral researcher at IMBB-FORTH, below the supervision of Dr. Poirazi.