Sub-symbolically managing pieces of symbolical functions for sorting

Area of Science:

• Artificial Intelligence
• Machine Learning
• Knowledge Representation

Background:

• Traditional AI systems often struggle with problems requiring both formal logic and pattern recognition.
• Gaps in formal theories limit the applicability of purely symbolic AI solutions.
• Integrating diverse knowledge types is crucial for advancing AI problem-solving.

Purpose of the Study:

• To develop a unified hybrid system capable of managing both symbolic and subsymbolic knowledge.
• To address limitations in formal theories by incorporating neural network modules.
• To create a flexible and adaptable AI framework for complex tasks.

Main Methods:

• A hybrid system architecture combining symbolic reasoning (e.g., formulas, rules) with neural modules.
• Training the entire system using a backpropagation learning algorithm for parameter updates.
• Utilizing a file sorting problem as a test-bed to evaluate system performance.

Main Results:

• The hybrid system successfully managed symbolic and subsymbolic knowledge uniformly.
• Neural modules effectively connected disparate pieces of symbolic knowledge.
• The system demonstrated capability in providing sorting move suggestions, integrating conventional sorter hints.

Conclusions:

• The proposed hybrid system offers a powerful approach to overcoming limitations in purely symbolic AI.
• This framework enables the uniform management of diverse knowledge types, enhancing problem-solving capabilities.
• The system's flexibility and adaptability show promise for various complex AI applications.