Research map / Optimization / Implicit bias and generalization
How Does Loss Function Affect Generalization Performance of Deep Learning? Application to Human Age Estimation
Ali Akbari, Muhammad Awais, Manijeh Bashar, Josef Kittler
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 |
|---|---|---|---|
| replace KL divergence with GJM loss in LDL | VGG age estimation, trained on BAG · avg over 5 test sets | MAE: 4.75 -> 4.30 | Improved |
| replace KL divergence with GJM loss in LDL | VGG age estimation, trained on SubBAG · avg over 5 test sets | MAE: 5.55 -> 5.08 | Improved |
| replace KL divergence with GJM loss in LDL | VGG age estimation, trained on MORPH · avg over 5 test sets | MAE: 7.24 -> 7.12 | Improved |
| replace KL divergence with GJM loss in LDL | ResNet50 age estimation, trained on SubBAG · avg over 5 test sets | MAE: 6.85 -> 5.90 | Improved |
| replace KL divergence with GJM loss | VGG age estimation · generalization error, avg over test sets | generalization error: 0.7038 -> 0.2037 | Improved |
| use label distribution learning (KL) instead of classification | VGG age estimation, trained on BAG · avg over 5 test sets | MAE: 5.10 -> 4.75 | Improved |
| vary GJM alpha from 0.5 to 0.2 (ablation) | GJM loss, VGG · FG-NET | MAE: 3.2 -> 3.7 | Hurt |
| vary GJM alpha from 0.5 to 0.8 (ablation) | GJM loss, VGG · FG-NET | MAE: 3.2 -> 4.8 | Hurt |
| vary GJM alpha from 0.5 to 0.4 (ablation) | GJM loss, VGG · FG-NET | MAE: 3.2 -> 3.2 | No clear effect |
Closest papers by meaning
- Mean-Variance Loss for Deep Age Estimation From a Face (CVPR 2018 · similarity 0.87)
- Penalizing Gradient Norm for Efficiently Improving Generalization in Deep Learning (ICML 2022 · similarity 0.86)
- Understanding Generalization and Optimization Performance of Deep CNNs (ICML 2018 · similarity 0.86)
- Understanding the Impact of Introducing Constraints at Inference Time on Generalization Error (ICML 2024 · similarity 0.85)
- On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima (ICLR 2017 · similarity 0.85)
- Generalizability of Neural Networks Minimizing Empirical Risk Based on Expressive Power (ICLR 2025 · similarity 0.84)
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.
