Holographic Declarative Memory: Distributional Semantics as the Architecture of Memory.

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Computational Neuroscience

Background:

• Modern machine learning excels at pattern recognition but struggles with high-level cognition.
• Symbolic cognitive architectures offer explainability but lack scalability for big data.
• Vector-symbolic architectures integrate symbolic reasoning with vector-based representations.

Purpose of the Study:

• To implement key cognitive architecture components and capabilities using vector-space operations.
• To bridge the gap between opaque neural networks and explainable symbolic models.
• To introduce Holographic Declarative Memory (HDM) as a scalable, vector-space cognitive model.

Main Methods:

• Utilizing distributional semantics to represent symbols as vectors in high-dimensional spaces.
• Implementing cognitive functions like learning, memory retrieval, and decision-making as algebraic operations.
• Developing the Holographic Declarative Memory (HDM) model based on vector-space principles.

Main Results:

• HDM successfully models primacy/recency effects in free recall and the fan effect in recognition.
• The model accounts for human probability judgments and performance in iterated decision tasks.
• HDM demonstrates a scalable and flexible approach to cognitive modeling.

Conclusions:

• Cognitive architectures can be effectively implemented as algebraic operations in vector spaces.
• HDM offers a unified framework bridging symbolic, quantum, and neural models of cognition.
• This approach provides an explanatory model for the internal representations of neural architectures.