Autonomous learning of sequential tasks: experiments and analyses

Area of Science:

• Artificial Intelligence
• Cognitive Science
• Machine Learning

Background:

• Hybrid learning models offer potential for integrating diverse knowledge representations.
• Bridging neural and symbolic learning is crucial for comprehensive AI systems.
• Sequential decision tasks require robust learning mechanisms for effective performance.

Purpose of the Study:

• To present the CLARION model, a novel hybrid learning system.
• To integrate neural, reinforcement, and symbolic learning paradigms.
• To investigate the model's efficacy in sequential decision-making.

Main Methods:

• Developed a two-level hybrid model named CLARION.
• Integrated neural, reinforcement, and symbolic learning components.
• Employed on-line, bottom-up learning from neural to symbolic representations.
• Utilized both procedural (neural) and declarative (symbolic) knowledge.

Main Results:

• The CLARION model successfully handled sequential decision tasks.
• Experimental analyses highlighted the model's advantages.
• Demonstrated effective integration of procedural and declarative knowledge.
• Showcased the benefits of bottom-up learning from neural to symbolic levels.

Conclusions:

• The CLARION model provides a synergistic approach to learning.
• Hybrid models can effectively combine different learning strategies.
• The model shows promise for complex sequential decision problems.