Intelligent Media Technology and Innovative Applications: Information-Theoretic Principles for Transcoding System Optimization

The rapid development of artificial intelligence technology is bringing profound changes and challenges to the media industry. Generative AI (AIGC), especially large models for text‑to‑image and text‑to‑video, have become hot research topics in the media field, demonstrating outstanding performance and huge potential across image, audio, video, and 3D domains.

The Future Media Network Collaborative Innovation Center of the School of Electronic Information and Electrical Engineering at Shanghai Jiao Tong University, together with the National Elite Engineer Academy, will co‑host a seminar titled “Intelligent Media Technology and Innovative Applications” on Friday, December 22, 2023, from 15:00 to 17:30, at the E‑Valley – Wuke Theater, 1st Floor, Building 4, Electrical Engineering Campus, Minhang, Shanghai.

Recommended Topic: Information-Theoretic Principles for Transcoding System Optimization

Speaker: Cai Chunlei – Team Leader of the Bilibili Multimedia Algorithms team.

Talk Introduction: The transcoding system of an internet video company is a large and complex pipeline composed of pre‑analysis, pre‑ and post‑processing, adaptive parameter decision, and codec modules. Joint optimization of the entire system requires massive effort in both algorithms and engineering. This talk explores whether a unified theoretical framework can be distilled from the myriad optimization methods—one that both explains the effectiveness of existing approaches and guides the design and optimization of practical transcoding systems.

Starting from information‑theoretic principles, the speaker will construct a concise theoretical framework and use it to derive the design rationale and optimization paths of current methods. The discussion will cover the evolution of mainstream transform‑and‑prediction coding frameworks, trends in traditional coding standards, the optimization principles behind the rapid rise of deep‑learning‑based coding, and predictions on how large‑model coding techniques may further improve compression ratios in the future.

A complete agenda will be provided during the event.

Experts from media platforms such as Xiaohongshu, as well as academic scholars, will gather to discuss topics including large models and generative intelligent media, intelligent optimization of audio‑video systems, and the expression and transmission of volumetric video.

The seminar aims to deeply examine the current state of technology, existing challenges, and future application scenarios, contributing to the further development of intelligent media technology. We look forward to your participation!