A scale-dependent notion of effective dimension

01/29/2020
by   Oksana Berezniuk, et al.
18

We introduce a notion of "effective dimension" of a statistical model based on the number of cubes of size 1/√(n) needed to cover the model space when endowed with the Fisher Information Matrix as metric, n being the number of observations. The number of observations fixes a natural scale or resolution. The effective dimension is then measured via the spectrum of the Fisher Information Matrix regularized using this natural scale.

READ FULL TEXT

page 1

page 2

page 3

page 4

research
03/10/2018

On a metric generalization of the tt-degrees and effective dimension theory

In this article, we study an analogue of tt-reducibility for points in c...
research
02/23/2022

On The Scale Dependence and Spacetime Dimension of the Internet with Causal Sets

A statistical measure of dimension is used to compute the effective aver...
research
10/30/2020

The power of quantum neural networks

Fault-tolerant quantum computers offer the promise of dramatically impro...
research
12/04/2019

Diffeological statistical models, the Fisher metric and probabilistic mappings

In this note I introduce a class of almost 2-integrable C^k-diffeologica...
research
10/24/2019

Wasserstein information matrix

We study the information matrix for statistical models by L^2-Wasserstei...
research
04/26/2021

Model-centric Data Manifold: the Data Through the Eyes of the Model

We discover that deep ReLU neural network classifiers can see a low-dime...
research
11/02/2022

Properties of the Concrete distribution

We examine properties of the Concrete (or Gumbel-softmax) distribution o...

Please sign up or login with your details

Forgot password? Click here to reset