Maryland MIRRORS/II: a connectionist model simulator.

Area of Science:

• Computer Science
• Biomedical Informatics
• Artificial Intelligence

Background:

• Developing and evaluating connectionist models (neural networks) is complex and time-intensive.
• Existing systems lack high-level, user-friendly interfaces for connectionist model construction.
• Biomedical applications require efficient and accessible tools for neural network development.

Purpose of the Study:

• To introduce Maryland MIRRORS/II, a software system designed to facilitate the construction of connectionist models.
• To highlight the unique features of Maryland MIRRORS/II that enhance usability in biomedicine.
• To demonstrate the practical application of the system in a biomedical context.

Main Methods:

• Design and implementation of the Maryland MIRRORS/II software system.
• Incorporation of a high-level nonprocedural language for model specification.
• Integration of a general-purpose event-handling mechanism and an indexed library of system resources.
• Utilizing error back propagation learning for a treatment selection example.

Main Results:

• Maryland MIRRORS/II provides a convenient and efficient platform for building connectionist models.
• The system's features, including its nonprocedural language and resource library, simplify model development.
• Demonstrated successful application in selecting appropriate medical treatments based on patient manifestations using error back propagation.

Conclusions:

• Maryland MIRRORS/II offers a significant advancement in the development of connectionist models, particularly for biomedical applications.
• The software system effectively addresses the challenges of complexity and time consumption in neural network construction.
• Maryland MIRRORS/II is a valuable tool for researchers and practitioners in biomedicine and related fields.