Conference

I will tell stories. About Lee's physics. About his insights that have shaped many of the ideas that we take for granted today. From the birth of Loop Quantum Gravity to the discreteness of physical space.

Machine learning has significantly improved the way scientists model and interpret large datasets across a broad range of the physical sciences; yet, its "black box" nature often limits our ability to trust and understand its results. Interpretable and explainable AI is ultimately required to realize the potential of machine-assisted scientific discovery. I will review efforts toward explainable AI focusing in particular in applications within the field of Astrophysics. I will present an explainable deep learning framework which combines model compression and information theory to achieve explainability. I will demonstrate its relevance to cosmological large-scale structures, such as dark matter halos and galaxies, as well as the cosmic microwave background, revealing new physical insights derived from these explainable AI models.

As we've seen at this workshop, exciting progress has recently been made in the study of neural networks by applying ideas and techniques from theoretical physics. In this talk, I will discuss a precise relation between quantum field theory and deep neural networks, the NN/QFT correspondence. In particular, I will go beyond the level of analogy by explicitly constructing the QFT corresponding to a class of networks encompassing both vanilla feedforward and recurrent architectures. The resulting theory closely resembles the well-studied O(N) vector model, in which the variance of the weight initializations plays the role of the 't Hooft coupling. In this framework, the Gaussian process approximation used in machine learning corresponds to a free field theory, and finite-width effects can be computed perturbatively in the ratio of depth to width, T/N. These provide corrections to the correlation length that controls the depth to which information can propagate through the network, and thereby sets the scale at which such networks are trainable by gradient descent. This analysis provides a non-perturbative description of networks at initialization, and opens several interesting avenues to the study of criticality in these models.

We will examine the idea of Bayesian updating as an inverse diffusion like process and its relation to the exact renormalisation group. In particular we will look at the role of Fisher Information, its metric and possible physical interpretations.

We show that Polchinski’s equation for exact renormalization group flow is equivalent to the optimal transport gradient flow of a field-theoretic relative entropy. This gives a surprising information-theoretic formulation of the exact renormalization group, expressed in the language of optimal transport. We will provide reviews of both the exact renormalization group, as well as the theory of optimal transportation. Our techniques generalize to other RG flow equations beyond Polchinski's. Moreover, we establish a connection between this more general class of RG flows and stochastic Langevin PDEs, enabling us to construct ML-based adaptive bridge samplers for lattice field theories. Finally, we will discuss forthcoming work on related methods to variationally approximate ground states of quantum field theories.

Feature learning - or the capacity of neural networks to adapt to the data during training - is often quoted as one of the fundamental reasons behind their unreasonable effectiveness. Yet, making mathematical sense of this seemingly clear intuition is still a largely open question. In this talk, I will discuss a simple setting where we can precisely characterise how features are learned by a two-layer neural network during the very first few steps of training, and how these features are essential for the network to efficiently generalise under limited availability of data.

We consider the problem of learning a generative model parametrized by a two-layer auto-encoder, and trained with online stochastic gradient descent, to sample from a high-dimensional data distribution with an underlying low-dimensional structure. We provide a tight asymptotic characterization of low-dimensional projections of the resulting generated density, and evidence how mode(l) collapse can arise. On the other hand, we discuss how in a case where the architectural bias is suited to the target density, these simple models can efficiently learn to sample from a binary Gaussian mixture target distribution.

As progress in AI hurtles forward at a speed seldom seen in the history of science, theorists who wish to gain a first-principles understanding of AI can be overwhelmed by the enormous number of papers, notational choices, and assumptions in the literature. I will make a pitch for developing a “Theoretical Minimum” for theoretical physicists aiming to study AI, with the goal of getting members of our community up to speed as quickly as possible with a suite of standard results whose validity can be checked by numerical experiments requiring only modest compute. In particular, this will require close collaboration between statistical physics, condensed matter physics, and high-energy physics, three communities that all have important perspectives to bring to the table but whose notation must be harmonized in order to be accessible to new researchers. I will focus my discussion on (a) the various approaches to the infinite-width limit, which seems like the best entry point for theoretical physicists who first encounter neural networks, and (b) the need for benchmark datasets from physics complex enough to capture aspects of natural-language data but which are nonetheless “calculable” from first-principles using tools of theoretical physics.

Diffusion models have shown remarkable success in generating high-dimensional data such as images and language – a feat only possible if data has strong underlying structure. Understanding deep generative models thus requires understanding the structure of the data they learn from. In particular, natural data is often composed of features organized hierarchically. In this talk, we will model this structure using probabilistic context-free grammars – tree-like generative models from linguistics. I will present a theory of denoising diffusion on this data, predicting a phase transition that governs the reconstruction of features at various hierarchical levels. I will show empirical evidence for it in both image and language diffusion models. I will then discuss how diffusion models learn these grammars, revealing a quantitative relationship between data correlations and the training set size needed to learn how to hierarchically compose new data. In particular, we predict a polynomial scaling of sample complexity with data dimension, providing a mechanism by which diffusion models avoid the curse of dimensionality. Additionally, this theory predicts that models trained on limited data generate outputs that are locally coherent but lack global consistency, an effect empirically confirmed across modalities. These results offer a new perspective on how generative models learn to become creative and compose novel data by progressively uncovering the latent hierarchical structure.