Research map / Architectures and efficiency / RNNs and normalization
Learning Anytime Predictions in Neural Networks via Adaptive Loss Balancing
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 constant loss weights with AdaLoss adaptive inverse-average-loss weights | ResANNs · CIFAR10, CIFAR100, SVHN | relative test error increase from OPT: 15~18% | Helped |
| Replace constant loss weights with AdaLoss adaptive inverse-average-loss weights | DenseANN169 · ILSVRC validation | top-1 error | Helped |
| Replace constant loss weights with AdaLoss adaptive inverse-average-loss weights | MSDNet32 · ILSVRC validation | top-1 error: 4.0e9 FLOPS; 6.6e9 total FLOPS | Helped |
| Use AdaLoss rather than CONST weights in the exponentially deepening ensemble | EANN · CIFAR100 | test error | Helped |
| Assemble exponentially deepening anytime networks instead of linear ensembles | EANN · ILSVRC validation | anytime accuracy | Helped |
| Use AdaLoss rather than CONST weights | small ANNs compared with larger CONST-weight ANNs · CIFAR100, CIFAR10, SVHN, ILSVRC | computation to reach an accuracy level: 50~100% | Helped |
| Give the final loss extra weight (ablation) | AdaLoss ANNs · ILSVRC validation | final top-1 error | No clear effect |
Closest papers by meaning
- Adaptive Neural Networks for Efficient Inference (ICML 2017 · similarity 0.88)
- Deep Learning meets Nonparametric Regression: Are Weight-Decayed DNNs Locally Adaptive? (ICLR 2023 · similarity 0.87)
- Towards Anytime Classification in Early-Exit Architectures by Enforcing Conditional Monotonicity (NeurIPS 2023 · similarity 0.87)
- Towards the Training of Deeper Predictive Coding Neural Networks (ICML 2026 · similarity 0.86)
- Wide Neural Networks Trained with Weight Decay Provably Exhibit Neural Collapse (ICLR 2025 · similarity 0.85)
- Why Do We Need Weight Decay in Modern Deep Learning? (NeurIPS 2024 · 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.