• P Lokeshwar Reddy
• Vandana Bhattacharjee

Zero-Shot Learning (ZSL) is a transformative machine learning approach that enables models to classify unseen classes without needing labeled data for those categories, making it highly valuable in data-scarce environments. This paper compares two significant ZSL methods: ZeroDiff and the Diffusion Classifier, exploring their strengths, weaknesses, and appropriate use cases. ZeroDiff is a Diffusion-based Generative ZSL model that excels at creating features for unseen classes, enhancing generalization through its dual-branch architecture, which integrates feature generation with semantic alignment. While effective, it requires substantial computational resources, making it less suitable for environments with limited hardware. In contrast, the Diffusion Classifier uses pretrained models to perform zero-shot classification by estimating class-conditional densities. Its ability to classify unseen data with minimal retraining makes it computationally efficient, but its performance heavily depends on the quality of the pretrained model, which can limit its versatility in handling a wide variety of unseen classes. The paper discusses the trade-offs between flexibility in feature generation and efficiency in classification and proposes the potential of hybrid models that combine both strengths. It also highlights key challenges, such as domain shifts and data scarcity, and suggests future research directions to improve model robustness, scalability, and adaptability to different domains.

"Effectiveness of Diffusion Models in Zero-Shot Learning", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.11, Issue 11, page no.f543-f554, November-2024, Available :http://www.jetir.org/papers/JETIR2411561.pdf

"Effectiveness of Diffusion Models in Zero-Shot Learning", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.11, Issue 11, page no. ppf543-f554, November-2024, Available at : http://www.jetir.org/papers/JETIR2411561.pdf