Synergistic information supports modality integration and flexible learning in neural networks solving multiple tasks

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Understanding cognition involves analyzing brain information processing modes.
• Synergistic information, encoded by neuron sets but not subsets, is vital for complex cognition.
• Key questions remain on how systems become synergistic and how information maps to artificial neural networks.

Purpose of the Study:

• Investigate how and why cognitive systems achieve high synergy.
• Analyze how informational states map onto artificial neural networks during learning.
• Explore the role of synergy in flexible and efficient learning.

Main Methods:

• Employed an information-decomposition framework.
• Investigated artificial neural networks performing cognitive tasks.
• Analyzed information dynamics during multi-task learning.

Main Results:

• Synergy increases as neural networks learn multiple diverse tasks.
• Performance in tasks requiring multi-source integration critically depends on synergistic neurons.
• Synergy facilitates combining information from multiple modalities.

Conclusions:

• Synergy is essential for flexible and efficient learning in artificial systems.
• Findings offer new insights into information-processing strategies in learning systems.
• General-purpose learning capacity relies on system information dynamics.