Autonomous Learning With High-Dimensional Computing Architecture Similar to Von Neumann's

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Artificial Intelligence

Background:

• Traditional computing relies on numbers and von Neumann architecture.
• Human and animal learning involve complex memory systems.
• Deep learning models excel at data-driven learning but lack biological plausibility.

Purpose of the Study:

• To propose a novel computing architecture for modeling biological learning.
• To bridge the gap between traditional computing, psychology, and neuroscience.
• To develop a theory of computing with vectors suitable for future technologies.

Main Methods:

• Modeling learning using high-dimensional vectors (D=10,000).
• Designing a computer architecture with vector-based operations and high-capacity vector memory.
• Drawing parallels with psychological models of working and long-term memory.
• Incorporating insights from neuroscience, specifically cerebellar cortex models.

Main Results:

• The proposed architecture mimics aspects of human and animal learning.
• It offers an alternative to deep learning with greater biological relevance.
• The model aligns with psychological theories of memory and learning.
• It provides a framework for understanding brain computation.

Conclusions:

• A theory of computing with vectors can elucidate brain computation.
• This architecture has potential applications in robotics and language processing.
• Future work requires mathematical theory development and large-scale experiments.
• The goal is to achieve brain-like material and energy efficiency in computation.