Dynamic Neural Network Structure: A Review for its Theories and Applications

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Neuroscience

Background:

• Static neural networks have limitations in accuracy, efficiency, and interpretability.
• Dynamic neural networks (DNNs) present advantages due to flexible structures and parameters.
• The evolution of broad learning systems (BLS) has expanded DNN applications beyond traditional deep learning (DL).

Purpose of the Study:

• To provide a comprehensive review of dynamic neural network (DNN) structures.
• To explore advancements in DNNs within deep learning (DL) and broad learning (BL).
• To discuss techniques for dynamic construction and inference, applications, and future research directions.

Main Methods:

• Literature review focusing on dynamic neural networks.
• Analysis of techniques for dynamic construction and inference in DNNs.
• Examination of DNN applications across diverse domains.

Main Results:

• DNNs demonstrate significant improvements in accuracy, efficiency, and interpretability compared to static networks.
• Rapid development in DNN structures is observed in both deep learning (DL) and broad learning (BL) domains.
• DNNs are applicable across a wide range of fields, with ongoing research addressing open issues.

Conclusions:

• Dynamic neural networks represent a significant advancement over static models.
• Further research into DNNs, particularly in DL and BL, holds promise for future innovations.
• This review serves as a valuable resource for researchers investigating DNNs and their applications.