scholarly journals Universality of Sine-Kernel for Wigner Matrices with a Small Gaussian Perturbation

2010 ◽  
Vol 15 (0) ◽  
pp. 526-604 ◽  
Author(s):  
Laszlo Erdos ◽  
Jose Ramirez ◽  
Benjamin Schlein ◽  
Horng-Tzer Yau
Keyword(s):  
Wigner Matrices ◽  
Gaussian Perturbation ◽  
Sine Kernel

Limiting Behavior of Eigenvectors of Large Wigner Matrices

2011 ◽  
Vol 146 (3) ◽  
pp. 519-549 ◽  
Author(s):  
Z. D. Bai ◽  
G. M. Pan
Keyword(s):  
Limiting Behavior ◽  
Wigner Matrices

scholarly journals Equipartition principle for Wigner matrices

2021 ◽  
Vol 9 ◽  
Author(s):  
Zhigang Bao ◽  
László Erdős ◽  
Kevin Schnelli
Keyword(s):  
High Precision ◽  
Quantum Systems ◽  
Wigner Matrices ◽  
Very High

Abstract We prove that the energy of any eigenvector of a sum of several independent large Wigner matrices is equally distributed among these matrices with very high precision. This shows a particularly strong microcanonical form of the equipartition principle for quantum systems whose components are modelled by Wigner matrices.

scholarly journals Local laws for polynomials of Wigner matrices

2020 ◽  
Vol 278 (12) ◽  
pp. 108507
Author(s):  
László Erdős ◽  
Torben Krüger ◽  
Yuriy Nemish
Keyword(s):  
Wigner Matrices

scholarly journals Regularized Training and Tight Certification for Randomized Smoothed Classifier with Provable Robustness

2020 ◽  
Vol 34 (04) ◽  
pp. 3858-3865
Author(s):  
Huijie Feng ◽  
Chunpeng Wu ◽  
Guoyang Chen ◽  
Weifeng Zhang ◽  
Yang Ning
Keyword(s):  
Neural Network ◽  
High Probability ◽  
Deep Neural Network ◽  
State Of The Art ◽  
Computationally Efficient ◽  
Base Classifier ◽  
Training Scheme ◽  
Adversarial Training ◽  
Gaussian Perturbation ◽  
Probabilistic Robustness

Recently smoothing deep neural network based classifiers via isotropic Gaussian perturbation is shown to be an effective and scalable way to provide state-of-the-art probabilistic robustness guarantee against ℓ2 norm bounded adversarial perturbations. However, how to train a good base classifier that is accurate and robust when smoothed has not been fully investigated. In this work, we derive a new regularized risk, in which the regularizer can adaptively encourage the accuracy and robustness of the smoothed counterpart when training the base classifier. It is computationally efficient and can be implemented in parallel with other empirical defense methods. We discuss how to implement it under both standard (non-adversarial) and adversarial training scheme. At the same time, we also design a new certification algorithm, which can leverage the regularization effect to provide tighter robustness lower bound that holds with high probability. Our extensive experimentation demonstrates the effectiveness of the proposed training and certification approaches on CIFAR-10 and ImageNet datasets.

Sign in / Sign up

Export Citation Format

Share Document