AI Paper Atlas

Research map / Optimization / Optimizers

A Unified Framework of Surrogate Loss by Refactoring and Interpolation

Lanlan Liu, Mingzhe Wang, Jia Deng

ECCV 2020 · Optimization › Optimizers · approach: Constrained and non-backprop training · 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
use UniLoss instead of MSE with best sigmaStacked Hourglass · MPII valPCKh: 95.74 -> 95.77No clear effect
use UniLoss instead of CEResNet-20 · CIFAR-10accuracy: 91.49 -> 91.64No clear effect
use UniLoss instead of CEResNet-20 · CIFAR-100accuracy: 65.90 -> 65.92No clear effect
use UniLoss instead of CE3-layer FC · MNIST zero vs non-zeroAP: 0.9989 -> 0.9988No clear effect
choose MSE sigma 1 instead of 0.1MSE heatmap loss · MPII valPCKh: 91.31 -> 95.71Improved
choose MSE sigma 3 instead of 10MSE heatmap loss · MPII valPCKh: 92.25 -> 95.74Improved
increase batch size from 8 to 128 (ablation)UniLoss · CIFAR-10accuracy: 87.89 -> 91.64Improved
increase batch size from 128 to 1024 (ablation)UniLoss · CIFAR-10accuracy: 91.64 -> 87.20Hurt
increase anchors from 5 to 16 (ablation)UniLoss · MPII valPCKh: 94.79 -> 95.77Improved
increase anchors from 5 to 16 (ablation)UniLoss · CIFAR-10accuracy: 91.55 -> 91.64No clear effect
increase anchors from 5 to 16 (ablation)UniLoss · MNIST zero vs non-zeroAP: 0.9988 -> 0.9988No clear effect
larger hamming distance from training configurations (ablation)interpolator · metric approximationrank correlation: 0.98 -> 0.15Hurt

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