LIMIT.Lab
-
Building multimodal AI foundation models with very limited resources!
AI foundation models are increasingly dominating various academic and industrial fields, yet the R&D of related technologies is controlled by limited institutions capable of managing extensive computational and data resources. To counter this dominance, there is a critical need for technologies that can develop practical AI foundation models using the standard computational and data resources. It is said that the scaling laws no longer provide the reliable roadmap for developing AI foundational models. Our community (LIMIT.Community) and the international lab (LIMIT.Lab) therefore aim to put in place exactly those technologies that permit the construction of {Vision, Vision-Language, Multimodal}AI foundational models even when compute and data are limited. Drawing on our members’ prior successes in (i) generative pre-training methods that can be applied horizontally across any modality with image, video, 3D, & audio, and (ii) high-quality AI models from extremely scarce data (including a single image), we have been committed to AI multimodal foundational models under very limited resources. As of 2025, LIMIT.Lab is composed primarily of international research teams from Japan, UK, and Germany. Through collaborative research projects and the workshop organization, we actively foster global exchange in the field of AI and related areas.
-
Post destination
Limited Resources, Unlimited Impact with Multimodal AI Models
-
Member
AIST, Science Tokyo, TUS
Oxford VGG, Cambridge
UTN FunAI Lab
UvA CVLab
Related Links