AI4Science: Opportunities and Challenges
06 Feb 2024, 15:30 — Room 322, DIBRIS , Via Dodecaneso 35
![Screenshot 2024-02-01 at 10.55.18 - [Azzurro Linea Font]](/media/images/IMG_3237_aoiwbt_HXsTi5w.width-800.jpg)
Speaker:
Luca Biggio — EPFL - Ecole polytechnique fédérale de Lausanne
Luca Biggio — EPFL - Ecole polytechnique fédérale de Lausanne
Abstract:
Interest in the intersection between Artificial Intelligence (AI) and Natural Sciences has grown rapidly in recent years. AI methods hold the promise of playing an increasingly important role in propelling scientific advancements across various disciplines such as biology, astrophysics and particle physics, to name a few. Concurrently, significant research efforts have been devoted to attaining a deeper understanding of the theoretical properties of AI models, examined as fundamental scientific entities. In this talk, I will delve into my research at the intersection of AI and Science, discussing both promising avenues and existing challenges within this evolving field. Initially, I will discuss various research works aimed at leveraging AI as a tool to support scientific discovery and accelerate computationally intensive numerical simulations. Next, I will present recent research highlighting distinct properties inherent in contemporary deep learning architectures, with a specific emphasis on Transformers. The primary focus will be on signal propagation, revealing some interesting inductive biases that Transformers impose on the routing of inputs through their layer hierarchy.
Interest in the intersection between Artificial Intelligence (AI) and Natural Sciences has grown rapidly in recent years. AI methods hold the promise of playing an increasingly important role in propelling scientific advancements across various disciplines such as biology, astrophysics and particle physics, to name a few. Concurrently, significant research efforts have been devoted to attaining a deeper understanding of the theoretical properties of AI models, examined as fundamental scientific entities. In this talk, I will delve into my research at the intersection of AI and Science, discussing both promising avenues and existing challenges within this evolving field. Initially, I will discuss various research works aimed at leveraging AI as a tool to support scientific discovery and accelerate computationally intensive numerical simulations. Next, I will present recent research highlighting distinct properties inherent in contemporary deep learning architectures, with a specific emphasis on Transformers. The primary focus will be on signal propagation, revealing some interesting inductive biases that Transformers impose on the routing of inputs through their layer hierarchy.
Bio:
Luca Biggio is an AI4Science fellow at EPFL. After graduating in Physics at the University of Genoa and obtaining a Master's degree in Machine Learning and Machine Intelligence at the University of Cambridge, he earned his PhD at ETHZ with a thesis on Transformers. His research lies at the intersection of AI and Science, with the goal of enhancing our comprehension of AI models and leveraging them to boost scientific discoveries in the Natural Sciences.
Luca Biggio is an AI4Science fellow at EPFL. After graduating in Physics at the University of Genoa and obtaining a Master's degree in Machine Learning and Machine Intelligence at the University of Cambridge, he earned his PhD at ETHZ with a thesis on Transformers. His research lies at the intersection of AI and Science, with the goal of enhancing our comprehension of AI models and leveraging them to boost scientific discoveries in the Natural Sciences.