Research map / Architectures and efficiency / RNNs and normalization
Betty: An Automatic Differentiation Library for Multilevel Optimization
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 |
|---|---|---|---|
| Use AID-FD instead of ITD-RMAD (ablation) | MWN · CIFAR-10 class imbalance, IF 200 | test accuracy: 68.91 -> 75.45 | Helped |
| Use AID-CG instead of ITD-RMAD (ablation) | MWN · CIFAR-10 class imbalance, IF 200 | test accuracy: 71.96 -> 66.23±1.88 | Hurt |
| Use AID-NMN instead of ITD-RMAD (ablation) | MWN · CIFAR-10 class imbalance, IF 200 | test accuracy: 71.96 -> 66.45±1.18 | Hurt |
| Increase AID-FD unrolling steps from 1 to 5 (ablation) | MWN · CIFAR-10 class imbalance, IF 200 | test accuracy: 75.45±0.63 -> 76.56±1.19 | Helped |
| Use mixed-precision training instead of full precision (ablation) | MWN with BERT-base · SST IF 20 | test accuracy: 89.99±0.38 -> 91.06±0.09 | Helped |
| Use full-precision training (ablation) | MWN with BERT-base · SST IF 20 | GPU memory: Out-of-memory | Hurt |
| Add label correction to the reweighting scheme (ablation) | WRENCH label correction and reweighting · WRENCH datasets | test accuracy: TREC 57.56±1.41 -> 66.76±1.31; AGNews 82.79±0.10 -> 83.16±0.20; IMDB 77.18±0.13 -> 77.80±0.26; SemEval 77.23±3.38 -> 84. | Helped |
| Add label correction to the reweighting scheme (ablation) | WRENCH label correction and reweighting · WRENCH YouTube | test accuracy: 91.60±0.75 -> 91.52±0.66 | No clear effect |
| Add a reweighting level to pretraining and finetuning (ablation) | OfficeHome domain adaptation · OfficeHome transfer directions | target-domain accuracy: Cl→Ar 65.43±0.36 -> 67.76±0.83; Ar→Pr 87.62±0.33 -> 88.53±0.42; Pr→Rw 77.43±0.41 -> 78.58±0.17; Rw→Cl 68.76±0.13 -> 69.7 | Helped |
| Increase DARTS unrolling steps from 1 to 3 (ablation) | DARTS reimplementation · CIFAR-10 | test accuracy: 97.39% -> 97.22% | Hurt |
| Enable mixed-precision training (ablation) | DARTS · DARTS memory analysis | GPU memory: 9867MiB -> 5759MiB | Helped |
| Enable mixed-precision training (ablation) | MWN with ResNet-50 · MWN memory analysis | GPU memory: 6817MiB -> 4397MiB | Helped |
Closest papers by meaning
- MetaBox-v2: A Unified Benchmark Platform for Meta-Black-Box Optimization (NeurIPS 2025 · similarity 0.88)
- Differentiation Through Black-Box Quadratic Programming Solvers (NeurIPS 2025 · similarity 0.87)
- Learning Provably Improves the Convergence of Gradient Descent (NeurIPS 2025 · similarity 0.86)
- SYMBOL: Generating Flexible Black-Box Optimizers through Symbolic Equation Learning (ICLR 2024 · similarity 0.86)
- Deep Distributed Optimization for Large-Scale Quadratic Programming (ICLR 2025 · similarity 0.86)
- Memory-Efficient Gradient Unrolling for Large-Scale Bi-level Optimization (NeurIPS 2024 · similarity 0.86)
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.