Key highlights:
- Sungkyunkwan University becomes the 32nd academic partner to adopt Theta EdgeCloud for cutting-edge AI research.
- Theta’s decentralized GPU platform will support research in vision-language modeling, domain adaptation, and robotic perception.
- A Samsung-backed AI paper from the university’s AIM Lab will be featured as a Highlight at CVPR 2025.
South Korean research lab partners with Theta to scale AI experimentation
Sungkyunkwan University’s AI & Media Lab (AIM Lab), under the leadership of Professor Sungeun Hong, has integrated Theta EdgeCloud into its computing infrastructure. This partnership makes the university the 32nd academic institution to tap into Theta Labs’ decentralized GPU platform, designed to support large-scale AI and machine learning research.
The AIM Lab, known for its contributions in multimodal learning and 3D vision, plans to use the platform for a wide range of tasks—spanning vision-language modeling to domain adaptation and privacy-conscious domain transfer. Theta EdgeCloud’s pay-as-you-go architecture provides scalable GPU power without the overhead typically associated with traditional centralized infrastructure. This setup is expected to help the lab train and deploy AI models with greater speed and lower cost.
Samsung-supported research earns international acclaim
The collaboration comes at a time when the AIM Lab is gaining broader recognition in the global AI research community. One of its latest projects, titled Question-Aware Gaussian Experts for Audio-Visual Question Answering, has been accepted as a Highlight Paper at CVPR 2025, an event known for its highly selective review process with an acceptance rate of just 2.98%. The study introduces QA-TIGER, a model that aligns key audio and visual frames based on user queries to improve accuracy in video-based question answering.
QA-TIGER is a smart AI model that answers questions about videos with both sound and visuals. It focuses on the most important moments in the video by using the question to guide what parts to look at. Instead of treating all frames equally, it selects and aligns the right audio and video segments using Gaussian attention and multiple expert models. This helps the AI give more accurate answers – Figure 3 shows examples of how QA-TIGER works for different types of questions (like “which instrument?”, “when does it play?”, etc.) on the MUSIC-AVQA dataset. It highlights how the model correctly finds and focuses on the right parts of the video and audio based on the question.
Another recent achievement is the publication of a second paper, Memory-Efficient Cross-Modal Attention for RGB-X Segmentation and Crowd Counting, in the journal Pattern Recognition. The work proposes a lightweight attention mechanism that reduces memory requirements for multimodal image segmentation, offering potential applications in fields like robotics and smart infrastructure. Both projects were supported by Samsung, which has a long-standing research affiliation with Sungkyunkwan University.
Growing network of academic adopters
By joining Theta’s ecosystem, Sungkyunkwan University now collaborates with top-tier institutions such as Stanford, Seoul National University, KAIST, and NTU Singapore. This expanding academic network is united by the goal of improving AI system performance and scalability through decentralized computing.
Professor Hong emphasized the lab’s intent to leverage Theta’s capabilities across multiple research areas, including 3D vision and robotic perception.
“Theta EdgeCloud is now the clear leader in decentralized GPU infrastructure for research institutions worldwide. We’re excited to get the computing flexibility and scale we need to push the boundaries of multimodal AI. From 3D vision to robot perception, Theta enables us to train and evaluate models more rapidly and cost-effectively. This will be game changing for our research lab.” —Professor Sungeun Hong, Director of the AI & Media Lab at SKKU
Mitch Liu, CEO and co-founder of Theta Labs, welcomed the university by noting the significance of its link to Samsung.
“We’re super proud to welcome Sungkyunkwan University with its strong research alliance with Samsung to the Theta family. As more Samsung-supported and globally respected research institutions join us, Theta EdgeCloud continues to set the standard for scalable, decentralized AI infrastructure. There is no doubt that Theta is becoming a household name in academia and research, and our reputation is now spreading by word of mouth and referrals.” —Mitch Liu, CEO and co-founder of Theta Labs
The bottom line
Theta EdgeCloud’s partnership with Sungkyunkwan University marks another milestone in its push to democratize AI research infrastructure. With backing from Samsung and a growing portfolio of influential academic adopters, Theta is establishing itself as a vital resource for research teams seeking both performance and affordability in AI experimentation.
Source:: Theta EdgeCloud Boosts Multimodal AI Research at Samsung-Linked Korean University