KAIST AI Will Use Theta EdgeCloud to Accelerate AI Model Training as First Academia Customer
Theta has already entered a new era in 2024 with the release of EdgeCloud in May. Now, we’re thrilled to announce a new partnership between Theta Labs and KAIST AI, marking a significant convergence of our EdgeCloud and cutting-edge academic research. This multi-year agreement will see KAIST AI, one of the world’s leading AI research institutions, utilizing Theta EdgeCloud to power their cutting-edge computer vision and natural language processing projects.
KAIST, often referred to as the “MIT of Korea,” has consistently ranked among the top institutions globally for machine learning research. Their decision to adopt Theta EdgeCloud speaks volumes about the platform’s real-world capabilities and potential advance AI infrastructure in a decentralized way.
At the heart of this collaboration is Professor Jaegul Choo, an Endowed Chair Professor at KAIST, whose team will be leveraging Theta EdgeCloud for their innovative work. One of their recent breakthroughs, StableVITON, showcases the power of AI in the fashion industry. This fine-tuned stable diffusion model can seamlessly superimpose clothing onto images of individuals, preserving their features and pose with remarkable accuracy.
As you may know, what sets Theta EdgeCloud apart is its unique hybrid architecture. As Professor Choo notes, “EdgeCloud is the only hybrid platform in the market where a particular AI job can be optimized for price, performance and depending on requirements routed to either high performing A100s/H100s cloud GPUs or to NVIDIA 3090s/4090s desktop GPUs as part of Theta’s distributed edge network.” This flexibility allows researchers to tailor their computing resources to the specific needs of each project, maximizing efficiency and cost-effectiveness.
The partnership doesn’t stop at StableVITON. KAIST AI and Theta Labs are already planning their next collaborative milestone: expanding the model to include new wearables, including more complex objects like hats. This GPU-intensive task will harness the full power of Theta EdgeCloud’s cluster of A100s and H100s, pushing the boundaries of what’s possible in AI-driven fashion technology.
For Theta, this partnership represents a major step into the academic research sector. Our EdgeCloud platform, powered by over 30,000 distributed edge nodes and cloud partners like Google Cloud and Amazon Web Services, offers over 80 PetaFLOPS of always available distributed GPU compute power. This makes it an ideal solution for the complex, resource-intensive tasks that cutting-edge AI research demands.
The implications of this research extend far beyond academia. As Mitch Liu, co-founder and CEO of Theta Labs, points out, “This unlocks a market opportunity of over 30,000 universities worldwide with multiple AI research groups and labs, spending billions of dollars every year for GPU-intensive AI tasks.”
You can try the Virtual Try-on Generative AI tool here, have fun!
