Generative AI Meets Semantic Communication: Evolution and Revolution of Communication Tasks

Eleonora Grassucci,Jihong Park,Sergio Barbarossa,Seong-Lyun Kim,Jinho Choi,Danilo Comminiello
2024-01-10
Abstract:While deep generative models are showing exciting abilities in computer vision and natural language processing, their adoption in communication frameworks is still far underestimated. These methods are demonstrated to evolve solutions to classic communication problems such as denoising, restoration, or compression. Nevertheless, generative models can unveil their real potential in semantic communication frameworks, in which the receiver is not asked to recover the sequence of bits used to encode the transmitted (semantic) message, but only to regenerate content that is semantically consistent with the transmitted message. Disclosing generative models capabilities in semantic communication paves the way for a paradigm shift with respect to conventional communication systems, which has great potential to reduce the amount of data traffic and offers a revolutionary versatility to novel tasks and applications that were not even conceivable a few years ago. In this paper, we present a unified perspective of deep generative models in semantic communication and we unveil their revolutionary role in future communication frameworks, enabling emerging applications and tasks. Finally, we analyze the challenges and opportunities to face to develop generative models specifically tailored for communication systems.
Computation and Language,Machine Learning
What problem does this paper attempt to address?
This paper discusses the application of generative artificial intelligence in semantic communication and proposes a unified perspective to understand the revolutionary role of deep generative models in the 6th generation (6G) communication system. Traditional communication systems mainly focus on the encoding, transmission, and recovery of information, while semantic communication goes further by aiming to generate content at the receiving end that is semantically equivalent to the transmitted information, without the need to fully recover the original bit sequence. Generative models have great potential in this field as they can reduce data traffic, increase system flexibility and adaptability, and support emerging tasks and applications. The paper mentions several common generative models, including Variational Autoencoders (VAEs), Flow-based models, Generative Adversarial Networks (GANs), and Diffusion Models (DDPMs), each with its own advantages and disadvantages, suitable for different semantic communication frameworks. The key lies in using semantic conditions to guide the generation process to ensure the generated content at the receiving end is semantically consistent with the transmitted information. The paper also discusses how these models can improve traditional communication tasks such as semantic compression beyond perceptual compression and modular architecture, as well as enable new applications such as channel-adaptive communication, multimodal semantic diversity, and emerging generative applications like content creation, multi-user communication, and network digital twins. In summary, this paper aims to demonstrate the potential of deep generative models in semantic communication and addresses the problem of how to utilize these models to achieve more efficient, flexible, and semantically accurate communication to meet the challenges and requirements of future communication systems.