91̽��

Abstract:

The remarkable performance of overparameterized deep neural networks (DNNs) must arise from an interplay between network architecture, training algorithms, and structure in the data. To disentangle these three components for supervised learning, we apply a Bayesian picture based on the functions expressed by a DNN. The prior over functions is determined by the network architecture, which we vary by exploiting a transition between ordered and chaotic regimes. For Boolean function classification, we approximate the likelihood using the error spectrum of functions on data. Combining this with the prior yields an accurate prediction for the posterior, measured for DNNs trained with stochastic gradient descent. This analysis shows that structured data, together with a specific Occam’s razor-like inductive bias towards (Kolmogorov) simple functions that exactly counteracts the exponential growth of the number of functions with complexity, is a key to the success of DNNs.

Abstract:

We show that Hubbard models with nearest-neighbor hopping and a nearest-neighbor hardcore constraint exhibit “maximal” Hilbert space fragmentation in many lattices of arbitrary dimension 𝑑. Focusing on the 𝑑 =1 rhombus chain and the 𝑑 =2 Lieb lattice, we demonstrate that the fragmentation is strong for all fillings in the thermodynamic limit, and explicitly construct all emergent integrals of motion, which include an extensive set of higher-form symmetries. Blockades consisting of frozen particles partition the system in real space, leading to anomalous dynamics. Our results are potentially relevant to optical lattices of dipolar and Rydberg-dressed atoms.