Scaling limits of wide neural networks with weight sharing: Gaussian process behavior, gradient independence, and neural tangent kernel derivation
G Yang - arXiv preprint arXiv:1902.04760, 2019 - arxiv.org
Several recent trends in machine learning theory and practice, from the design of state-of-
the-art Gaussian Process to the convergence analysis of deep neural nets (DNNs) under …
the-art Gaussian Process to the convergence analysis of deep neural nets (DNNs) under …
Entropy and mutual information in models of deep neural networks
We examine a class of stochastic deep learning models with a tractable method to compute
information-theoretic quantities. Our contributions are three-fold:(i) We show how entropies …
information-theoretic quantities. Our contributions are three-fold:(i) We show how entropies …
Convergence of smoothed empirical measures with applications to entropy estimation
This paper studies convergence of empirical measures smoothed by a Gaussian kernel.
Specifically, consider approximating P* N σ, for N σ=△ N (0, σ 2 I d), by P̑ n* N σ under …
Specifically, consider approximating P* N σ, for N σ=△ N (0, σ 2 I d), by P̑ n* N σ under …
Generalization error of generalized linear models in high dimensions
M Emami, M Sahraee-Ardakan… - International …, 2020 - proceedings.mlr.press
At the heart of machine learning lies the question of generalizability of learned rules over
previously unseen data. While over-parameterized models based on neural networks are …
previously unseen data. While over-parameterized models based on neural networks are …
Inference with deep generative priors in high dimensions
P Pandit, M Sahraee-Ardakan… - IEEE Journal on …, 2020 - ieeexplore.ieee.org
Deep generative priors offer powerful models for complex-structured data, such as images,
audio, and text. Using these priors in inverse problems typically requires estimating the input …
audio, and text. Using these priors in inverse problems typically requires estimating the input …
Inference in deep networks in high dimensions
AK Fletcher, S Rangan… - 2018 IEEE International …, 2018 - ieeexplore.ieee.org
Deep generative networks provide a powerful tool for modeling complex data in a wide
range of applications. In inverse problems that use these networks as generative priors on …
range of applications. In inverse problems that use these networks as generative priors on …
The spiked matrix model with generative priors
Using a low-dimensional parametrization of signals is a generic and powerful way to
enhance performance in signal processing and statistical inference. A very popular and …
enhance performance in signal processing and statistical inference. A very popular and …
Random linear estimation with rotationally-invariant designs: Asymptotics at high temperature
We study estimation in the linear model, in a Bayesian setting where has an entrywise iid
prior and the design is rotationally-invariant in law. In the large system limit as dimension …
prior and the design is rotationally-invariant in law. In the large system limit as dimension …
The mutual information in random linear estimation beyond iid matrices
There has been definite progress recently in proving the variational single-letter formula
given by the heuristic replica method for various estimation problems. In particular, the …
given by the heuristic replica method for various estimation problems. In particular, the …
Information-theoretic limits for the matrix tensor product
G Reeves - IEEE Journal on Selected Areas in Information …, 2020 - ieeexplore.ieee.org
This article studies a high-dimensional inference problem involving the matrix tensor product
of random matrices. This problem generalizes a number of contemporary data science …
of random matrices. This problem generalizes a number of contemporary data science …