Implicit regularization in hierarchical tensor factorization and deep convolutional neural networks

N Razin, A Maman, N Cohen - International Conference on …, 2022 - proceedings.mlr.press
In the pursuit of explaining implicit regularization in deep learning, prominent focus was
given to matrix and tensor factorizations, which correspond to simplified neural networks. It …

The effect of smooth parametrizations on nonconvex optimization landscapes

E Levin, J Kileel, N Boumal - Mathematical Programming, 2024 - Springer
We develop new tools to study landscapes in nonconvex optimization. Given one
optimization problem, we pair it with another by smoothly parametrizing the domain. This is …

[图书][B] Metric algebraic geometry

P Breiding, K Kohn, B Sturmfels - 2024 - library.oapen.org
Metric algebraic geometry combines concepts from algebraic geometry and differential
geometry. Building on classical foundations, it offers practical tools for the 21st century …

Critical points and convergence analysis of generative deep linear networks trained with Bures-Wasserstein loss

P Bréchet, K Papagiannouli, J An… - … on Machine Learning, 2023 - proceedings.mlr.press
We consider a deep matrix factorization model of covariance matrices trained with the Bures-
Wasserstein distance. While recent works have made advances in the study of the …

On the minimal algebraic complexity of the rank-one approximation problem for general inner products

K Kozhasov, A Muniz, Y Qi, L Sodomaco - arXiv preprint arXiv:2309.15105, 2023 - arxiv.org
We study the algebraic complexity of Euclidean distance minimization from a generic tensor
to a variety of rank-one tensors. The Euclidean Distance (ED) degree of the Segre-Veronese …

Understanding deep learning via notions of rank

N Razin - arXiv preprint arXiv:2408.02111, 2024 - arxiv.org
Despite the extreme popularity of deep learning in science and industry, its formal
understanding is limited. This thesis puts forth notions of rank as key for developing a theory …

Function space and critical points of linear convolutional networks

K Kohn, G Montúfar, V Shahverdi, M Trager - SIAM Journal on Applied …, 2024 - SIAM
We study the geometry of linear networks with one-dimensional convolutional layers. The
function spaces of these networks can be identified with semialgebraic families of …

Side effects of learning from low-dimensional data embedded in a Euclidean space

J He, R Tsai, R Ward - Research in the Mathematical Sciences, 2023 - Springer
The low-dimensional manifold hypothesis posits that the data found in many applications,
such as those involving natural images, lie (approximately) on low-dimensional manifolds …

The geometry of the deep linear network

G Menon - arXiv preprint arXiv:2411.09004, 2024 - arxiv.org
This article provides an expository account of training dynamics in the Deep Linear Network
(DLN) from the perspective of the geometric theory of dynamical systems. Rigorous results …

Geometry of lightning self-attention: Identifiability and dimension

NW Henry, GL Marchetti, K Kohn - arXiv preprint arXiv:2408.17221, 2024 - arxiv.org
We consider function spaces defined by self-attention networks without normalization, and
theoretically analyze their geometry. Since these networks are polynomial, we rely on tools …