MIPT and AIRI: Biologically plausible computational memory model for artificial intelligence systems

The main articles are:

• Artificial Intelligence (Russian market)
• Artificial Intelligence (Global Market)

2022: Creating a biologically plausible memory model for artificial intelligence systems

On May 11, 2022, Russian scientists from the Institute of Artificial Intelligence AIRI and MIPT announced the creation of a biologically plausible memory model for artificial intelligence systems with internal motivation. A scientific article was published in the journal Brain Informatics.

Cognitive an agent is one who program learns to interact with the world on her own and learn from her mistakes by performing a specific task. The basis of the agent is the architecture of, algorithms among other things, neural network which helps him execute the developer's instructions.

Human nerve cell

In everyday life, we regularly encounter the results of the work of machine learning methods and artificial intelligence. Over the past decade, success in this direction has been associated with the training of deep neural networks (Artificial Neural Networks, ANN), built on the basis of an artificial neuron model. The researchers also highlight spike neural networks (SNNs) built around a spike neuron model that is closer to a biological neuron. Artificial neural networks exchange real numbers, and these ones exchange adhesions, unit events that occur at a certain time, repeating the work of the nervous system.

Artificial neural networks are more common due to the simplicity of the neuron model used, and the architecture of graphics accelerators is suitable for related calculations. They use all the neurons they contain to transmit information, while spike neural networks emulate the work of the brain of an animal or a person - they use only neurons that are active at a particular moment in time, which provides significant resource savings when learning and using them. In addition, it is spike neural networks, biologically plausible and hybrid models and methods of AI training that are considered more promising in terms of progress in understanding the principles of the human brain through the possibilities of their use in cognitive sciences. At the heart of such developments is the use of a model of a pyramidal neuron, which makes up the bulk of neurons in the human cortex and learns faster than an artificial neuron.

Alexander Panov, Head of the Group "Neural Integration" of the Institute of Artificial Intelligence AIRI, Head of the Laboratory of Cognitive Dynamic Systems of the Center for Cognitive Modeling MIPT

Researchers of the Neurosimvol Integration group of the AIRI Institute of Artificial Intelligence and students of the Moscow Institute of Physics and Technology have created, according to them, the first biologically plausible computational model of agent memory in Russia, which is able to effectively operate in an unfamiliar environment under the influence of an external reinforcement signal. For example, navigate and search for resources in labyrinths and rooms.

The developed agent model can operate with abstractions of states and actions. This means that he is able to perform complex actions on the basis of simple operations already known to him. For example, by learning to search the door indoors, the agent will be able to use this skill to solve more complex tasks, while most programs require the creation of an instruction for each specific task. In addition to external motivation (rewards for a successfully performed action), the agent developed by the scientific group also has an internal one. This makes his behavior more complex and autonomous. Internal motivation provides meaningful behavior in the absence of an external reinforcement signal. This means that such an agent will be able not only to look for a solution to the problem, like most standard programs, but also to study the world around him.

The article was carried out as part of a long-term fundamental study at the junction of computer and cognitive sciences, which brings scientists closer to creating more independent artificial intelligence systems and a better understanding of the principles of human and animal brains. The construction of such large hybrid biologically plausible models and their subsequent testing in complex simulation environments is a little investigated direction. Projects like these help cognitive researchers test theories and hypotheses, and the model created could be one example of how a whole set of models from neuroscience can be connected together to make the work of AI systems more similar to that of the human brain. In addition, the biological plausibility of the structure of such an agent will require less computing power and make AI more economical.

"The architecture developed by our group is a complex of interesting ideas from the field of neuroscience and cognitive research. These are implementations of hierarchical presentation of information, internal motivation and prediction of the consequences of events. Our main achievement is to teach such a comprehensive system to act stably and consistently in the environment. Our model gives researchers from all over the world the opportunity to create even more complex systems that even better mimic the work of the human brain and psyche, " says Alexander Panov, Head of the Neurosimilar Integration Group at the AIRI Institute of Artificial Intelligence, Head of the Laboratory of Cognitive Dynamic Systems at the MIPT Center for Cognitive Modeling

The code of the cognitive agent is publicly available, anyone who wants to researcher can use it.