KTH GAIN – Generative AI for Next-Generation Science
The GAIN platform aims to build on KTH’s strengths in scientific computing to establish broad leadership in applying generative AI methods in high-performance computing environments. Our particular focus is achieving impact in high-profile scientific and societal challenges.
One of the strongest trends in AI research is the move towards generative networks that can produce complex models (e.g. a folded protein or chat interaction) instead of mere classifiers. These models require extreme-scale training in terms of both data and computational resources. As we run out of natural training data, the field increasingly relies on adding physics-aware components to models and generating new synthetic training data computationally.
KTH hosts some of the world’s most cited researchers in scientific computing who also lead international collaborations, e.g. with RIKEN. Moreover, KTH has been the driving force in establishing the new Swedish computational infrastructure with a significantly stronger AI focus, and we have been able to create leading KTH environments that span both computing and applications, e.g. as part of Science for Life Laboratory.
Furthermore, this platform integrates some of the leading researchers in AI for science, including groundbreaking contributions in a wide range of fields (e.g. climate, chemistry, materials, fluid mechanics, medicine, etc
Our vision is to make KTH leading in developing and training large foundation models in collaboration with international partners. These models are then fine-tuned for specific applications and combined with interactive data analytics to integrate how scientists generate, interact with and draw scientific conclusions from data.
We aim to develop an AI for Science platform at KTH to leverage significant infrastructure investments, EU-funded Centers-of-Excellence, and funding from Knut and Alice Wallenberg Foundation and national strategic research areas.
Contact
