Research map / Optimization / Optimizers
Towards Gradient-based Bilevel Optimization with Non-convex Followers and Beyond
Risheng Liu, Yaohua Liu, Shangzhi Zeng, Jin Zhang
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 |
|---|---|---|---|
| use IAPTT-GM (initialization auxiliary + pessimistic truncation) instead of RHG | bilevel gradient method · miniImageNet 5-way 1-shot | test accuracy (%): 50.54 -> 56.69 | Improved |
| use IAPTT-GM instead of RHG | bilevel gradient method · tieredImageNet 5-way 5-shot | test accuracy (%): 75.20 -> 75.85 | Improved |
| remove pessimistic trajectory truncation (IA-GM) (ablation) | IAPTT-GM · non-convex toy BLO | convergence | Improved |
| add pessimistic trajectory truncation (ablation) | IAPTT-GM · non-convex toy BLO | backward iterations / speed | Improved |
| use IAPTT-GM instead of BDA | bilevel gradient method · toy BLO | runtime (s): 16.800 -> 10.835 | Improved |
| use IAPTT-GM instead of RHG | bilevel gradient method · toy BLO | runtime (s): 9.990 -> 10.835 | Hurt |
| add initialization auxiliary variable (ablation) | IAPTT-GM · non-convex toy BLO | convergence from distant initial points | Improved |
| add Nesterov acceleration to IA-GM under lower-level convexity (ablation) | IA-GM(A) · convex BLO and hyper-cleaning | convergence speed / loss | Improved |
| use IAPTT-GM on data hyper-cleaning with two FC layers | IAPTT-GM · MNIST/FashionMNIST | accuracy and F1 | Improved |
Closest papers by meaning
- An Accelerated Gradient Method for Convex Smooth Simple Bilevel Optimization (NeurIPS 2024 · similarity 0.89)
- Conditional Gradient Methods with Standard LMO for Stochastic Simple Bilevel Optimization (NeurIPS 2025 · similarity 0.88)
- A Single-Loop Gradient Algorithm for Pessimistic Bilevel Optimization via Smooth Approximation (NeurIPS 2025 · similarity 0.87)
- Unlocking Global Optimality in Bilevel Optimization: A Pilot Study (ICLR 2025 · similarity 0.87)
- Penalty-based Methods for Simple Bilevel Optimization under Hölderian Error Bounds (NeurIPS 2024 · similarity 0.87)
- Tuning-Free Bilevel Optimization: New Algorithms and Convergence Analysis (ICLR 2025 · similarity 0.87)
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.
