Summary: New computational neuroscience research sheds light on how the brain’s cognitive abilities develop and could help shape new AI research.

Source: University of Montreal

New research introduces a new neurocomputational model of the human brain that sheds light on how the brain develops complex cognitive abilities and advances neural artificial intelligence research.

Published on September 19, the study was carried out by an international team of researchers from the Paris Institute Pasteur and Sorbonne University, CHU Sainte-Justine, Mila – Quebec Institute of Artificial Intelligence and Université de Montréal.

The model who made the cover of the magazine Proceedings of the National Academy of Sciences of the United States of America (PNAS), describes neural development at three hierarchical levels of information processing:

  • The first sensorimotor level examines how the inner workings of the brain learn patterns from perception and associate them with action;
  • The cognitive level examines how the brain integrates these patterns;
  • Finally, the level of consciousness deals with how the brain separates from the outside world and manages learned patterns (through memory).

The team’s research model provides insights into the underlying mechanisms of cognition because it focuses on the interaction between two basic types of learning: Hebbian learning, which is based on statistical regularity (i.e., repetition)—or, as neuropsychologist Donald Hebb put it, “nerves that fire together, wire together”—and reinforcement learning. , associated with reward and dopamine neurotransmission.

The model addresses three tasks of increasing complexity at those levels, from visual recognition to conscious awareness. Each time he introduced a new core method that allowed the team to make progress.

The results highlight two fundamental mechanisms for developing multi-level cognitive capabilities in biological neural networks:

  • Synaptic epigenesis, Hebbian learning at the local scale and reinforcement learning at the global scale;
  • and self-organized dynamics, with spontaneous activity and a balanced excitatory/inhibitory ratio of neurons.
This shows the brain
The model addresses three tasks of increasing complexity at those levels, from visual recognition to conscious awareness. The image is in the public domain.

“Our model shows how the integration of neuro-AI can highlight biological mechanisms and cognitive architecture, which can lead to next-generation development of artificial intelligence and eventually lead to artificial consciousness,” said team member Guillaume Dumas, computational psychiatry. Assistant Professor said. UdeM, and Principal Investigator at the CHU Sainte-Justine Research Center.

Getting to this point may require integrating the social dimension of cognition, he said. The researchers are now exploring the integration of biological and social dimensions in human cognition. The team previously pioneered the first simulation of two whole-brain interactions.

The convergence of future mathematical models in biological and social reality will continue to shed light on the underlying mechanisms of cognition, the team believes, but the only known system with advanced social consciousness will help provide a unique bridge to artificial intelligence. Brain.

About this computational neuroscience research news

Author: Julie Gazelle
Source: University of Montreal
Contact: Julie Gazelle – University of Montreal
Image: The image is in the public domain.

Preliminary study: Open Access.
Multilevel development of cognitive skills in artificial neural networks” by Guillaume Dumas et al. PNAS


Draft

watch out

This shows an old woman riding a bicycle.

Multilevel development of cognitive skills in artificial neural networks

Several neural mechanisms have been proposed to account for the formation of cognitive abilities in postnatal interactions with the physical and sociocultural environment.

Here, we introduce a three-level computational model of information management and cognitive capabilities. We provide minimal architectural requirements for building these standards and how the requirements affect their performance and connectivity.

The first sensorimotor stage handles local unconscious processing, here during a visual classification task. The second level, or cognitive level, integrates the information from several local processors through long-distance connections at the global level and integrates them in a global, but still unconscious way. The third and highest level of knowledge controls the information globally and consciously. It is based on the Global Neuronal Workspace (GNW) theory and is called the level of consciousness.

We use monitoring operations and evolution operations, respectively, to challenge the second and third steps. The results highlight the importance of epigenesis in selecting and stabilizing synapses at local and global scales to enable the network to resolve the first two functions.

Globally, although there is a temporal delay between perception and reward, dopamine appears to be important for the proper delivery of credit allocation. Third, in the absence of sensory neurons, the presence of interneurons would be necessary to maintain a self-sustaining representation in the GNW.

Finally, when balanced spontaneous internalization facilitates epigenesis both locally and globally, a balanced stimulatory/inhibitory ratio enhances its performance. We talk about the plausibility of the model in both neurodevelopment and artificial intelligence.

Leave a Reply

Your email address will not be published.