AI Paper Atlas

Research map / Learning theory and algorithms / Kernels and regression theory

Deep Equals Shallow for ReLU Networks in Kernel Regimes

Alberto Bietti, Francis Bach

ICLR 2021 · Learning theory and algorithms › Kernels and regression theory · approach: Neural tangent kernel analysis · 9 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
Increase ReLU random-feature network depthFully connected ReLU RF kernels · Sphere; asymptotic eigenvalue decayeigenvalue decay exponent: k^{-d-2}No clear effect
Increase ReLU network depthFully connected ReLU NTK kernels · Sphere; asymptotic eigenvalue decayeigenvalue decay exponent: k^{-d}No clear effect
Use step activations and compose the random-neuron kernel with depthStep-activation random-neuron kernels · Sphere; asymptotic eigenvalue decayeigenvalue decay exponent: ν = 1/2^{L-1}Improved
Use shallow instead of three-layer random ReLU featuresRandom-feature regression · Synthetic f₂*expected squared errorImproved
Increase RF kernel depth from 2 to 3 layersRF kernel ridge regression · MNIST testaccuracy: 98.60 ± 0.03 -> 98.67 ± 0.03No clear effect
Increase RF kernel depth from 2 to 3 layersRF kernel ridge regression · Fashion-MNIST testaccuracy: 90.75 ± 0.11 -> 90.87 ± 0.16No clear effect
Increase NTK depth from 2 to 3 layersNTK kernel ridge regression · MNIST testaccuracy: 98.49 ± 0.02 -> 98.53 ± 0.02No clear effect
Increase NTK depth from 2 to 3 layersNTK kernel ridge regression · Fashion-MNIST testaccuracy: 90.65 ± 0.07 -> 90.62 ± 0.08No clear effect
Use RF rather than NTK kernelsKernel ridge regression · Fashion-MNIST testaccuracy: 90.87 ± 0.16 vs 90.62 ± 0.08Improved

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