Mathematical Foundations of AI
Mathematical Foundations of AI
Applications are closed
Embedded Files
Nordic Summer Evening, 1899-1900, Richard Bergh (Göteborgs Konstmuseum)
"Is AI doomed to become an engineering-only field, or is it possible to build a science of AI?"
The workshop will explore the following topics:
Transformers: these architectures are at the core of Large Language Models and are rapidly revolutionising all fields of ML. Despite theoretical efforts, approaches remain scattered, and we lack a unifying framework. This theme aims to unify these perspectives on how to study transformers.
Diffusion Models/Flow Matching: these models are the state of the art for generative tasks. Theoretical progress in this field is proceeding at an unprecedented pace, making results from just one or two years ago obsolete. The goal is to assess the current state of research and describe the open questions that lie ahead.
Associative Memories: models of associative memory have recently returned as a core topic in AI research, given their analogies with aspects of learning and memorization in Transformers and Diffusion Models. In this theme, we aim to reconcile classical and recent results to repurpose them to understand these modern technologies.
A Day in March, Lofoten, 1899, Anna Boberg (Göteborgs Konstmuseum)
Invited talks
Invited talks
Program
Program
The program is available below, abstracts of talks are displayed here.
Day 1 - June 8th
Day 1 - June 8th
📅 Day 1 — Diffusion Models and Associative Memories
📅 Day 1 — Diffusion Models and Associative Memories
8:30–9:30 — Welcome Fika and Registration
9:30–10:30 — Luca Ambrogioni: How Out-of-Equilibrium Phase Transitions can Seed Pattern Formation in Trained Diffusion Models
10:30–11:00 — Chenxiao Ma (Contributed Talk): The interplay between data structure and imbalance in the learning dynamics of diffusion models
11:00–11:30 — Break
11:30–12:30 — Alberto Fachechi: Dense L-Directional Hopfield Networks: Merging Modularity and Higher-Order Interactions for Pattern Disentanglement
12:30–12:45 — Discussion
12:45–14:00 — Lunch
14:00–15:00 — Matteo Negri: Detecting Creativity Geometrically: Hopfield networks as a Lens on Generative AI
15:00–15:30 — Vincenzo Schimmenti (Contributed Talk): Equilibria of Modern Hopfield networks
15:30–16:00 — Afternoon Fika
16:00–17:00 — Beatrice Achilli: Statistical Physics of Generative Diffusion
17:00–17:30 — Enrico Ventura (Contributed Talk): Emergence of Distortions in High-Dimensional Guided Diffusion Models
17:30–17:45 — Discussion
Day 2 - June 9th
Day 2 - June 9th
📅 Day 2 — Geometric Deep Learning and Transformers
📅 Day 2 — Geometric Deep Learning and Transformers
9:00–10:00 — Morning Fika
10:00–11:00 — Daniel Persson: TBA
11:00–11:30 — Break
11:30–12:30 — Giovanni Marchetti: Algebraic Geometry of Deep Learning
12:30–12:45 — Discussion
12:45–14:00 — Lunch
14:00–15:00 — Freya Behrens: Model Systems for attention, structure and memory in Transformers
15:00–15:30 — Oskar Allerbo (Contributed Talk): Is supervised learning really that different from unsupervised?
15:30–16:00 — Afternoon Fika
16:00–17:00 — Peter Súkeník: One token to rule them all? Sink vs. diagonal patterns as mechanisms for attention switch and oversmoothing prevention
17:30–17:45 — Discussion
19:30 — Speakers' dinner
Day 3 - June 10th
Day 3 - June 10th
📅 Day 3 — Transformers
📅 Day 3 — Transformers
8:30–9:30 — Morning Fika
9:30–10:30 — Alessio Giorlandino: From Initialisation to Factual Recall: Two Studies on Transformers
10:30–11:00 — Alessandro Zambon (Contributed Talk): Sampling at intermediate temperatures is optimal for training large language models in protein structure prediction
11:00–11:30 — Break
11:30–12:30 — Yedi Zhang: Training Dynamics of In-Context Learning in Linear Attention
12:20–13:00 — Discussion and conclusion
Organisers
Organisers
Sponsor
Sponsor
Page updated
Google Sites
Report abuse