Research map / Optimization / Optimizers
Adaptive Methods for Nonconvex 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 |
|---|---|---|---|
| increase Adam epsilon from 1e-8 to 1e-3 (limit adaptivity) | ADAM · MNIST deep autoencoder | test loss: 4.36 -> 1.88 | Helped |
| increase Adam epsilon from 1e-8 to 1e-3 | ADAM · IWSLT'15 En-Vi | BLEU: 27.92 -> 28.28 | Helped |
| increase Adam epsilon from 1e-8 to 1e-3 at LR 1e-3 | ADAM · CIFAR-10 ResNet50 | test accuracy: 92.59 -> 91.74 | Hurt |
| replace Adam's multiplicative v update with Yogi's additive sign-based update | ADAM (eps=1e-3) · MNIST deep autoencoder | test loss: 1.88 -> 1.36 | Helped |
| use Yogi instead of Adam+Noam schedule | Transformer base · IWSLT'15 En-Vi | BLEU: 28.84 -> 29.27 | Helped |
| use Yogi instead of Adam+Noam+checkpoint averaging | Transformer base · WMT'14 En-De | BLEU: 27.3 vs 27.2 | No clear effect |
| use Yogi instead of SGD+momentum | ResNet50 · CIFAR-10 | test accuracy: 93.03 -> 93.90 | Helped |
| use Yogi instead of SGD+momentum | DenseNet-40 · CIFAR-10 | test accuracy: 94.76 vs 94.38 | Hurt |
| use Yogi default instead of tuned RMSProp | Inception-ResNet-v2 · ImageNet | top-1 error: 21.63% -> 21.14% | Helped |
| lower Yogi beta2 from 0.999 to 0.9 (ablation) | YOGI · MNIST deep autoencoder | test loss: 1.36 -> 1.70 | Hurt |
Closest papers by meaning
- On the Convergence of A Class of Adam-Type Algorithms for Non-Convex Optimization (ICLR 2019 · similarity 0.91)
- PAGE: A Simple and Optimal Probabilistic Gradient Estimator for Nonconvex Optimization (ICML 2021 · similarity 0.89)
- Adam with model exponential moving average is effective for nonconvex optimization (NeurIPS 2024 · similarity 0.88)
- Non-asymptotic Analysis of Biased Adaptive Stochastic Approximation (NeurIPS 2024 · similarity 0.88)
- Adaptive Methods through the Lens of SDEs: Theoretical Insights on the Role of Noise (ICLR 2025 · similarity 0.88)
- A Comprehensive Framework for Analyzing the Convergence of Adam: Bridging the Gap with SGD (ICML 2025 · similarity 0.88)
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.