Under the grant “Brain and Knowledge: From Natural Intelligence to Artificial Intelligence”, new neural network memory models similar to human memory were created at Moscow State University, the rector of the university, Academician of the Russian Academy of Sciences Viktor Sadovnichy told socialbites.ca about this.
“Analogs based on standardly trained recurrent networks and transformers cannot solve these problems effectively. The memory model we developed allows us to train algorithms to trigger “long-term memory,” which was not previously possible for programs. Our model allows us to find dependencies between events widely separated in time and store them in memory. The level of development corresponds to the level of world achievements in this field,” Sadovnichy explained.
According to the rector, the implementation of the project will help scientists understand how exactly information is encoded in the human nervous system and learn to transfer the biological principles of encoding conceptual information into neuromorphic artificial intelligence systems.
“We have also created an artificial synapse based on a film of semiconductor nanocrystals. It allows you to encode information in a series of pulses and also has short-term and long-term memory. The results obtained reveal new approaches to creating photovoltaic structures that could be the basis of future neuromorphic systems.” said.
Learn more about other developments in the native Silicon Valley, Russia’s largest university, methods of using the latest supercomputer and the joint Russia-China project. report Sadovnichy “socialbites.ca”.
Formerly Rector of Moscow State University Sadovnichy named Four applications of the latest supercomputer.