Iterative Deep Neighborhood: A Deep Learning Model Which Involves Both Input Data Points and Their Neighbors.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Deep Learning

Background:

• Deep learning models excel at automatic feature extraction from large datasets.
• Current deep learning models often ignore contextual information from neighboring data points, limiting performance.

Purpose of the Study:

• To demonstrate that incorporating neighboring data points significantly boosts deep learning performance.
• To introduce a novel deep learning model and iterative algorithm that leverage neighborhood information.

Main Methods:

• Designed a deep learning model accepting both data instances and neighbors' classification responses.
• Developed an iterative algorithm (Iterative Deep Neighborhood - IDN) for updating neighbors based on deep representations and model parameters.

Main Results:

• The proposed Iterative Deep Neighborhood (IDN) algorithm outperforms state-of-the-art deep learning models.
• Demonstrated significant performance improvements across diverse tasks including image classification, text sentiment analysis, and property price prediction.

Conclusions:

• Neighboring data points provide critical insights for enhancing deep learning classification.
• The IDN algorithm offers a powerful approach to improve deep learning model performance by utilizing contextual data relationships.