Research map / Optimization / Optimizers
Will Bilevel Optimizers Benefit from Loops
Kaiyi Ji, Mingrui Liu, Yingbin Liang, Lei Ying
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 |
|---|---|---|---|
| Add a large inner loop (N-loop) instead of using small N (ablation) | AID-BiO · Bilevel optimization, epsilon-accurate stationary point | MV(epsilon): \mathcal{\widetilde O}(\kappa^6\epsilon^{-1}) -> \mathcal{\widetilde O}(\kappa^4\epsilon^{-1}) | Improved |
| Add a large inner loop (N-loop) instead of using small N (ablation) | AID-BiO · Bilevel optimization, epsilon-accurate stationary point | Gc(epsilon): \mathcal{\widetilde O}(\kappa^6\epsilon^{-1}) -> \mathcal{\widetilde O}(\kappa^5\epsilon^{-1}) | Improved |
| Add a large Q loop for the Hessian-inverse-vector approximation (ablation) | AID-BiO with N=1 · Bilevel optimization, epsilon-accurate stationary point | Gc(epsilon): \mathcal{\widetilde O}(\kappa^6\epsilon^{-1}) -> \mathcal{\widetilde O}(\kappa^5\epsilon^{-1}) | Improved |
| Add a large Q loop for the Hessian-inverse-vector approximation (ablation) | AID-BiO with N=1 · Bilevel optimization, epsilon-accurate stationary point | MV(epsilon): \mathcal{\widetilde O}(\kappa^6\epsilon^{-1}) -> \mathcal{\widetilde O}(\kappa^6\epsilon^{-1}) | No clear effect |
| Use N=20 rather than N=1 (ablation) | ITD-BiO · Hyper-representation validation loss at k=500 | validation loss: 0.04 -> 0.004 | Improved |
| Use N=20 rather than N=1 (ablation) | ITD-BiO · Hyper-representation validation loss at k=1000 | validation loss: 0.04 -> 0.004 | Improved |
| Use N=20 rather than N=1 (ablation) | AID-BiO · MNIST hyperparameter optimization | training and test loss versus running time | Improved |
| Increase Q while keeping N fixed (ablation) | AID-BiO · MNIST hyperparameter optimization | convergence performance | No clear effect |
Closest papers by meaning
- Tuning-Free Bilevel Optimization: New Algorithms and Convergence Analysis (ICLR 2025 · similarity 0.88)
- A Nearly Optimal Single Loop Algorithm for Stochastic Bilevel Optimization under Unbounded Smoothness (ICML 2024 · similarity 0.88)
- SPABA: A Single-Loop and Probabilistic Stochastic Bilevel Algorithm Achieving Optimal Sample Complexity (ICML 2024 · similarity 0.86)
- A Single-Loop Gradient Algorithm for Pessimistic Bilevel Optimization via Smooth Approximation (NeurIPS 2025 · similarity 0.85)
- A Single-Loop First-Order Algorithm for Linearly Constrained Bilevel Optimization (NeurIPS 2025 · similarity 0.85)
- Unlocking Global Optimality in Bilevel Optimization: A Pilot Study (ICLR 2025 · similarity 0.85)
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.
