[Paper Review] Generative Artificial Intelligence: A Systematic Review and Applications

Sandeep Singh Sengar, Affan Bin Hasan+2 more|arXiv (Cornell University)|May 17, 2024

Artificial Intelligence in Healthcare26 citations

TL;DR

This paper surveys recent Generative AI techniques (GANs, VAEs, diffusion, transformers) and their applications across image, video, and language tasks, discussing datasets, metrics, challenges, and responsible AI considerations.

ABSTRACT

In recent years, the study of artificial intelligence (AI) has undergone a paradigm shift. This has been propelled by the groundbreaking capabilities of generative models both in supervised and unsupervised learning scenarios. Generative AI has shown state-of-the-art performance in solving perplexing real-world conundrums in fields such as image translation, medical diagnostics, textual imagery fusion, natural language processing, and beyond. This paper documents the systematic review and analysis of recent advancements and techniques in Generative AI with a detailed discussion of their applications including application-specific models. Indeed, the major impact that generative AI has made to date, has been in language generation with the development of large language models, in the field of image translation and several other interdisciplinary applications of generative AI. Moreover, the primary contribution of this paper lies in its coherent synthesis of the latest advancements in these areas, seamlessly weaving together contemporary breakthroughs in the field. Particularly, how it shares an exploration of the future trajectory for generative AI. In conclusion, the paper ends with a discussion of Responsible AI principles, and the necessary ethical considerations for the sustainability and growth of these generative models.

Motivation & Objective

• Summarize state-of-the-art generative AI techniques and architectures.
• Synthesize applications of generative models in image translation, video synthesis, and NLP.
• Compare datasets and evaluation metrics used to benchmark GenAI methods.
• Highlight challenges, opportunities, and future directions in responsible AI for GenAI.

Proposed method

• Conduct a targeted literature review (2012–2023) focused on Generative AI techniques and applications.
• Categorize models into GANs, transformers, VAEs, and diffusion models across sections.
• Discuss foundational issues (training stability, mode collapse) and subsequent improvements (W-GAN, LS-GAN, etc.).
• Review application domains with representative datasets and evaluation metrics (FID, KID, RMSE, SSIM, PSNR, LPIPS).
• Address ethical considerations and Responsible AI principles for GenAI.

Experimental results

Research questions

• RQ1 What are the major generative AI techniques and how have they evolved from 2012 to 2023?
• RQ2 How are GANs, VAEs, diffusion models, and transformers applied across image, video, and language tasks?
• RQ3 What datasets and metrics are commonly used to benchmark generative models, and what are the ethical considerations for deploying GenAI?
• RQ4 What are the key challenges and future directions in responsible GenAI development?

Key findings

• GANs addressed training divergence and mode collapse; improvements include W-GAN and LS-GAN.
• Transformers enabled powerful sequence modeling and foundational NLP models (e.g., GPT, BERT).
• VAEs provide probabilistic latent representations and have variants like denoising autoencoders; Bicycle GANs show balance of diversity and realism.
• Diffusion models and normalizing flows offer strong generative capabilities with iterative refinement.
• Applications span image translation (medical and satellite imagery), video synthesis (talking-head and expression-driven generation), and text-to-image and molecule generation.
• The paper emphasizes Responsible AI principles and ethical considerations for sustainable GenAI growth.