Advanced School on Foundation Models for Scientific Discovery | (smr 4088)
Starts 10 Jul 2025
Ends 18 Jul 2025
Central European Time
ICTP
Kastler Lecture Hall (AGH)
Strada Costiera, 11
I - 34151 Trieste (Italy)
The call for applications is open, please click on "Apply here" on the left, in order to submit your application
Foundation models, defined as deep learning (DL) models trained on extensive collections of unlabeled data and capable of adapting to various tasks with minimal fine-tuning, have emerged as pivotal tools for driving scientific discovery in different research fields (e.g. Molecular chemistry, Bioinformatics). These models address very well the common limitations of experimental scientific data, such as the scarcity of annotated data and the presence of data noise. The importance of Foundation models lies in their ability to process and analyse vast amounts of structured and unstructured data, uncovering hidden patterns that can lead to new scientific hypotheses and discoveries. Moreover, the universal approach of these tools facilitates knowledge transfer across diverse scientific fields.
We are excited to announce the second Advanced School in Applied ML at ICTP, which will be focused on "Foundation Models for Scientific Discovery". The school will take place in Trieste, Italy from July 10 to July 18, 2025. This intensive, interdisciplinary advanced school is designed for early-career researchers and PhD students eager to explore the transformative potential of foundation models in driving scientific discovery. It aims to bridge the gap between theoretical deep learning concepts and the application of foundation models to cutting-edge scientific problems. There will be a special focus on scalability, sustainability, and advanced HPC methods for optimizing deep learning pipelines.
Lecturers from top-ranked research centers/universities and big tech companies (e.g. IBM, Nvidia) will cover the following key topics:
Fundamentals of Foundation Models: Theory and Practice
Transformers for tabular data
Applications in Climate Science,Astrophysics, Material Science and Drug Discovery
Uncertainty Quantification (UQ) methods ( to assess and incorporate uncertainty in predictions made by AI models)
Explainable and trustworthy AI for Scientific research
Multimodal Learning
Each topic will be complemented by the necessary tools to create efficient, scalable, and portable ML pipelines that exploit the capabilities of modern HPC infrastructures.
30 participants will be selected with attention to creating a cohort with balanced representation of countries (particularly, least developed countries), gender and under-represented groups.
This school will be the second activity in the field of applied Machine Learning co-organised by the brand-new ICOMP (link: https://www.ictp.it/home/consortium-scientific-computing). This initiative, co-organized by ICTP alongside SISSA/TSDS, IBM and AREA, marks the first training activity co-sponsored under the ICTP-IBM partnership.
Research abstracts: In the application form, all applicants are requested to submit a brief research abstract for a poster presentation. A limited number of abstracts will be selected for the poster session. Please use ICTP templates available for download here and below under 'Material'.
Grants: A limited number of grants are available to support the attendance of selected participants, with priority given to participants from developing countries. There is no registration fee.
Jean Barbier (ICTP), Alberto Cazzaniga (AREA Science Park), Serafina Di Gioia (ICTP), Ivan Girotto (ICTP), Teodoro Laino (IBM), Roberto Trotta (SISSA), Local Organiser: Serafina Di Gioia (ICTP)