Foundation model adaptation
Title: Foundation model adaptation
DNr: Berzelius-2024-229
Project Type: LiU Berzelius
Principal Investigator: Kevin Smith <>
Affiliation: Kungliga Tekniska högskolan
Duration: 2024-06-26 – 2024-11-01
Classification: 10207


Foundation models have emerged as versatile tools for various tasks, yet adapting them is often computationally intensive and prone to overfitting. Existing methods for model adaptation, while aiming to ease this process, can introduce additional computational costs. This project aims to develop a new, lightweight adaptation method for large foundation models that enhances performance while being more efficient than current approaches. This is impactful for real-world transfer learning applications such as medical image analysis and mobile edge devices, where maximizing performance with minimal cost is crucial.