Engineering Innovations: How AI Is Changing Images and Video

Why It Matters

By cutting the size and energy cost of image and video data, AI‑driven compression accelerates remote healthcare, immersive media, and autonomous operations, delivering both economic and environmental benefits.

Key Takeaways

• AI-driven compression can outperform traditional JPEG/H.265 rates significantly
• Machine-oriented compression discards human‑irrelevant data for efficiency significantly
• Model pruning and quantization shrink AI models for edge devices
• Split inference reduces bandwidth by transmitting intermediate features only
• Faster visual data transmission enables remote healthcare, AR/VR, and automation

Summary

Engineering Innovations' podcast explores how AI is reshaping visual data compression. Professor Maggie Zu explains that traditional lossy codecs like JPEG and H.265 rely on fixed transform parameters, limiting adaptability and efficiency as video resolutions and formats proliferate. AI‑driven learned compressors use deep neural networks to automatically discover optimal representations, achieving higher compression ratios while maintaining visual quality.

Zu highlights two emerging fronts: designing compression for machines rather than humans, and shrinking the AI models themselves. In machine‑oriented scenarios—such as factory‑floor video sensors or billions of surveillance cameras—only task‑relevant features need to be retained, dramatically cutting bitrates. Model pruning and quantization further reduce computational load, enabling deployment on edge devices.

Concrete examples include split‑inference architectures that send intermediate feature maps from sensors to cloud servers, and faster transmission of medical imaging (MRI, CT) that can support remote diagnostics. These advances promise lower energy consumption, reduced storage costs, and new capabilities for AR/VR, autonomous systems, and telehealth.

Overall, AI‑enhanced compression could become a foundational technology for the data‑intensive future, unlocking bandwidth‑constrained applications while mitigating environmental impact.

Original Description

AI and Visual Data Compression: Insights from Purdue ECE Professor Maggie Zhu

In this episode of Engineering Innovations, host Kristin Malavenda explores the fascinating world of visual data compression with Professor Maggie Zhu of Purdue University's Elmore Family School of Electrical and Computer Engineering. They discuss how AI is revolutionizing the efficiency of image and video compression, the challenges and opportunities it presents, and how these advancements impact industries ranging from healthcare to entertainment. Tune in to learn about the latest research and future possibilities in AI-driven compression.

00:00 Introduction to Engineering Innovations Podcast

01:05 Understanding Visual Data Compression

03:19 Traditional vs. AI-Based Compression Methods

05:59 Challenges and Opportunities with AI in Compression

07:53 Designing Compression for Machines

11:43 Impact of Compression on Mobile and Cloud Systems

13:15 Compressing AI Models: Techniques and Benefits

15:13 Future Applications and Industry Impact

18:37 Personal Journey and Insights

26:37 Conclusion and Final Thoughts

Purdue ECE: http://engineering.purdue.edu/ECE

Facebook: http://facebook.com/purdueece

X: https://x.com/PurdueECE

Instagram: https://instagram.com/purdue.ece

LinkedIn: https://www.linkedin.com/company/purdue-ece

Purdue University's Elmore Family School of Electrical and Computer Engineering, founded in 1888, is one of the largest ECE departments in the nation and is consistently ranked among the best in the country.