AI Paper Atlas

Research map / Optimization / Optimizers

On the distance between two neural networks and the stability of learning

Jeremy Bernstein, Arash Vahdat, Yisong Yue, Ming-Yu Liu

NeurIPS 2020 · Optimization › Optimizers · approach: Constrained and non-backprop training · 16 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
use Fromage instead of SGDFromage · CIFAR-10 train lossloss: (1.5±0.2)×10^-4 -> (2.5±0.5)×10^-5Improved
use Fromage instead of AdamFromage · CIFAR-10 train lossloss: (6±3)×10^-5 -> (2.5±0.5)×10^-5Improved
use Fromage instead of AdamFromage · GAN CIFAR-10FID: 23.7 -> 16Improved
use Fromage instead of AdamFromage · ImageNet train lossloss: 2.02 -> 2.001No clear effect
use Fromage instead of AdamFromage · Transformer Wikitext-2 trainperplexity: 36.8 -> 66.1Hurt
use Fromage instead of SGDFromage · Transformer Wikitext-2 trainperplexity: 150.0 -> 66.1Improved
use Fromage instead of SGDFromage · GAN CIFAR-10FID: 34 -> 16Improved
use Fromage instead of off-the-shelf optimisers on very deep MLPs without skip connectionsFromage · MNIST MLP depth over 25training accuracyImproved
remove 1/sqrt(1+eta^2) prefactor (LARS-like update) (ablation)Fromage · spectrally normalised cGANnumerical stabilityHurt

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