Research map / Adversarial robustness and security / Black-box and transfer attacks
Generalizable Symbolic Optimizer Learning
Xiaotian Song, Peng Zeng, Yanan Sun, Andy Song
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 |
|---|---|---|---|
| learn symbolic optimizer with SOL instead of Adam | ConvNet · CIFAR-10 | accuracy (%): 67.15% -> 68.75% | Improved |
| learn symbolic optimizer with SOL instead of SymbolicL2O | ResNet-18 · CIFAR-10 | accuracy (%): 93.45% -> 93.16% | Hurt |
| learn symbolic optimizer with SOL instead of SymbolicL2O | ResNet-18 · CIFAR-10 | search time (h): 5.65h -> 1.60h | Improved |
| use SOL optimizer instead of AdamW, transferred from ResNet18 | LLaMA-7B · Alpaca fine-tuning | validation error: 0.93 -> 0.91 | Improved |
| use SOL instead of parametric L2LGD2 optimizer | ResNet50 · GPU memory | memory (MB): 24,541MB -> 3,169MB | Improved |
| use SOL optimizer instead of PGD | adversarial attack on seven defense models · RobustBench CIFAR-10 | attack success rate (%): 38.67% -> 39.45% (Carmon2019) | Improved |
| use SOL optimizer instead of APGD | adversarial attack · RobustBench CIFAR-10 | attack success rate (%): 39.06% -> 39.45% (Carmon2019); tie 55.08% on Wong2020 | No clear effect |
| use SOL optimizer instead of Adam | GAT and JK-Net · Cora/Citeseer/PubMed | accuracy (%): 82.48% -> 83.10% (GAT Cora) | Improved |
| use SOL optimizer instead of AdamW | BERT-base · GLUE (5 datasets) | score: 61.01 -> 63.14 (CoLA) | Improved |
| use SOL instead of parametric L2LGD2 for training loss | MNISTNET · MNISTNET training loss | convergence speed | Mixed |
Closest papers by meaning
- SYMBOL: Generating Flexible Black-Box Optimizers through Symbolic Equation Learning (ICLR 2024 · similarity 0.87)
- Adam-mini: Use Fewer Learning Rates To Gain More (ICLR 2025 · similarity 0.84)
- A Stable Whitening Optimizer for Efficient Neural Network Training (NeurIPS 2025 · similarity 0.84)
- MADA: Meta-Adaptive Optimizers Through Hyper-Gradient Descent (ICML 2024 · similarity 0.84)
- Investigation into Training Dynamics of Learned Optimizers (Student Abstract) (AAAI 2024 · similarity 0.84)
- SWAN: SGD with Normalization and Whitening Enables Stateless LLM Training (ICML 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.
