How deep is the brain? The shallow brain hypothesis

Area of Science:

• Computational neuroscience
• Artificial intelligence
• Neurobiology

Background:

• Deep learning and predictive coding models predominantly use hierarchical inference.
• Neurobiological evidence shows direct subcortical connections to all cortical areas, challenging purely hierarchical models.

Purpose of the Study:

• To propose the shallow brain hypothesis, integrating hierarchical cortical processing with parallel subcortical contributions.
• To highlight computational principles potentially missed by current deep learning architectures.

Main Methods:

• Conceptual analysis of neuroanatomy and computational architectures.
• Comparison of proposed shallow brain architecture with deep hierarchical models.

Main Results:

• Current hierarchical models may miss essential brain computational principles.
• A shallow architecture leverages cortical microcircuits and thalamo-cortical loops.

Conclusions:

• The shallow brain hypothesis offers a more biologically plausible and potentially more efficient computational framework.
• This model better explains the mammalian brain's fast and flexible capabilities compared to deep hierarchical structures.