A Review of Generalized Zero-Shot Learning Methods

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computer Vision

Background:

• Generalized zero-shot learning (GZSL) addresses the challenge of classifying data samples when some output classes are not seen during supervised training.
• GZSL utilizes semantic information from both seen (source) and unseen (target) classes to bridge the knowledge gap.

Purpose of the Study:

• To provide a comprehensive review of existing Generalized Zero-Shot Learning (GZSL) methods.
• To offer a structured categorization of GZSL approaches and discuss representative techniques.
• To identify research gaps and suggest future research directions in GZSL.

Main Methods:

• The review presents an overview of GZSL, including its inherent problems and challenges.
• A hierarchical categorization of GZSL methods is introduced.
• Representative methods within each category are discussed and analyzed.

Main Results:

• The paper categorizes various GZSL methods, offering a structured overview of the field.
• It discusses benchmark datasets and real-world applications of GZSL.
• Key research gaps and potential future research avenues are highlighted.

Conclusions:

• This review provides a structured overview of the GZSL landscape, categorizing existing methods and discussing their strengths and weaknesses.
• It highlights the importance of semantic information in bridging the gap between seen and unseen classes.
• The paper concludes by identifying critical research gaps and outlining promising directions for future GZSL investigations.