Research map / Optimization / Optimizers
AutoLoss: Learning Discrete Schedule for Alternate 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 AutoLoss scheduling instead of dense grid search over the L1 coefficient | d-ary quadratic regression · d-ary quadratic regression (test) | MSE: .086 -> .070 | Helped |
| Use AutoLoss scheduling instead of dense grid search over the L1 coefficient | MLP classification · MLP classification (test) | classification error: .093 -> .088 | Helped |
| Use AutoLoss scheduling rather than the vanilla alternating schedule | DCGAN · MNIST | inception score: 8.6307 -> 9.0549 | Helped |
| Use AutoLoss scheduling rather than the best reported fixed-ratio baseline | DCGAN · MNIST | inception score: 9.0232 -> 9.0549 | Helped |
| Use AutoLoss scheduling rather than fine-tuned multi-task training | multi-task NMT · German-to-English NMT (validation) | PPL: 3.64 -> 3.54 | Helped |
| Drop normalized gradient magnitude from the controller features (ablation) | AutoLoss · Regression (feature ablation) | MSE: .070 -> .086 | Hurt |
| Drop loss values from the controller features (ablation) | AutoLoss · Regression (feature ablation) | MSE: .070 -> .101 | Hurt |
| Drop validation metrics from the controller features (ablation) | AutoLoss · Regression (feature ablation) | MSE: .070 -> .085 | Hurt |
| Transfer the trained controller to another data distribution | MLP classification · MLP classification (transfer, Dataset 2) | MSE: .1035 -> .1016 | Helped |
| Transfer the trained controller to another data distribution | MLP classification · MLP classification (transfer, Dataset 3) | MSE: .1022 -> .0997 | Helped |
Closest papers by meaning
- Delve into the Applicability of Advanced Optimizers for Multi-Task Learning (NeurIPS 2026 · similarity 0.86)
- Towards AutoAI: Optimizing a Machine Learning System with Black-box and Differentiable Components (ICML 2024 · similarity 0.84)
- MetaOptimize: A Framework for Optimizing Step Sizes and Other Meta-parameters (ICML 2025 · similarity 0.84)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.84)
- PSMGD: Periodic Stochastic Multi-Gradient Descent for Fast Multi-Objective Optimization (AAAI 2025 · similarity 0.84)
- AutoOpt: A Dataset and a Unified Framework for Automating Optimization Problem Solving (NeurIPS 2025 · similarity 0.83)
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.