FAU’s video-compression technology portfolio has been acquired by Dolby Laboratories, underscoring the global impact of its research. The portfolio includes 683 intellectual property assets, 250 granted patents and 73 U.S. patents, including 16 U.S. patents essential to the international VVC/H.266 standard. The technology enables video to be delivered using about 50% less data without sacrificing quality, advancing streaming, broadcasting, defense communications, autonomous systems, robotics and AI worldwide.

Halide cathode materials are reshaping all-solid-state lithium batteries (ASSLBs) by enabling multi-electron reactions that dramatically boost energy density. A new review traces their evolution from liquid to solid-state systems, spotlighting single-phase electrode designs that eliminate inactive components and the rational design of multi-electron cathode materials. Looking ahead, it highlights AI-driven material discovery and integrated architectures to overcome energy challenges, charting a course toward cheaper, higher-energy ASSLBs.

Shaping electromagnetic waves at will is a long-standing goal in modern science and engineering. Programmable metasurfaces offer a promising route, but existing schemes are often too simple and fixed for real-time, task-oriented, and diverse control. A team from Zhejiang University has now developed a deep-learning-enabled temporal sequencing framework that maps target illusion patterns to modulation sequences, enabling rewritable and customizable electromagnetic illusions in experiments. This work is published in National Science Review.

Researchers at Aalto University in Finland, together with international partners, have developed the most accurate model yet of how humans read. The new model uses reinforcement learning, a type of AI used in robotics, to explain—and recreate—the choices readers make as they move through text.

Eye AI is moving fast, but will it work for people living at high altitude? In a study accepted by Science Bulletin, Dr. Wu Yuan’s team found that leading ophthalmic foundation models can lose accuracy across high-altitude retinal image cohorts. The team developed MIXFound, a lightweight way to combine several models and make AI diagnosis more robust in underrepresented settings.