Research map / Optimization / Optimizers
A Unified Framework of Surrogate Loss by Refactoring and Interpolation
Lanlan Liu, Mingzhe Wang, Jia Deng
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use UniLoss instead of MSE with best sigma | Stacked Hourglass · MPII val | PCKh: 95.74 -> 95.77 | No clear effect |
| use UniLoss instead of CE | ResNet-20 · CIFAR-10 | accuracy: 91.49 -> 91.64 | No clear effect |
| use UniLoss instead of CE | ResNet-20 · CIFAR-100 | accuracy: 65.90 -> 65.92 | No clear effect |
| use UniLoss instead of CE | 3-layer FC · MNIST zero vs non-zero | AP: 0.9989 -> 0.9988 | No clear effect |
| choose MSE sigma 1 instead of 0.1 | MSE heatmap loss · MPII val | PCKh: 91.31 -> 95.71 | Improved |
| choose MSE sigma 3 instead of 10 | MSE heatmap loss · MPII val | PCKh: 92.25 -> 95.74 | Improved |
| increase batch size from 8 to 128 (ablation) | UniLoss · CIFAR-10 | accuracy: 87.89 -> 91.64 | Improved |
| increase batch size from 128 to 1024 (ablation) | UniLoss · CIFAR-10 | accuracy: 91.64 -> 87.20 | Hurt |
| increase anchors from 5 to 16 (ablation) | UniLoss · MPII val | PCKh: 94.79 -> 95.77 | Improved |
| increase anchors from 5 to 16 (ablation) | UniLoss · CIFAR-10 | accuracy: 91.55 -> 91.64 | No clear effect |
| increase anchors from 5 to 16 (ablation) | UniLoss · MNIST zero vs non-zero | AP: 0.9988 -> 0.9988 | No clear effect |
| larger hamming distance from training configurations (ablation) | interpolator · metric approximation | rank correlation: 0.98 -> 0.15 | Hurt |
Closest papers by meaning
- Relational Surrogate Loss Learning (ICLR 2022 · similarity 0.87)
- On the Dynamics Under the Unhinged Loss and Beyond (JMLR 2023 · similarity 0.86)
- On Compressing Deep Models by Low Rank and Sparse Decomposition (CVPR 2017 · similarity 0.86)
- Machine Unlearning under Overparameterization (NeurIPS 2025 · similarity 0.85)
- Unified Gradient-Based Machine Unlearning with Remain Geometry Enhancement (NeurIPS 2024 · similarity 0.85)
- Preserving Deep Representations in One-Shot Pruning: A Hessian-Free Second-Order Optimization Framework (ICLR 2025 · similarity 0.85)
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.
