AI Paper Atlas

Research map / Learning theory and algorithms / Matrix and tensor methods

Non-Gaussian Tensor Programs

Eugene Golikov, Greg Yang

NeurIPS 2022 · Learning theory and algorithms › Matrix and tensor methods · approach: Tensor network decompositions · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
sample input/output (vector-like) weights from Rademacher instead of Gaussian with equal variance1-hidden-layer tied-weight network f(xi)=U^T phi(U xi)/n · width to infinity, phi(x)=x*1[|x|<=1/2]network output f(1)Hurt
sample vector-like weights from +-1 instead of standard GaussianNNGP of f=V phi(U xi)/sqrt(n) with indicator phi · infinite-width GP limitlimiting Gaussian process kernelHurt
swap Gaussian for non-Gaussian matrix-like (hidden) weights with the same mean and varianceTensor Programs / wide neural networks · infinite-width limitlimiting training behaviorNo clear effect
swap Gaussian for non-Gaussian initial vectors in a programTensor Programs · infinite-width limitprogram limitMixed
use a different initialization distribution (mean 0, variance 1) for abc-parametrization weightsabc-parametrization of MLPs · feature learning/kernel/stable/nontrivial regimesregime classificationNo clear effect
add bias parameters to the MLPPrinciple 1 universality claim for biasless MLP · wide MLP initializationvalidity of universality principleHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy