Bilevel Optimization: Convergence Analysis and Enhanced Design

Authors: Kaiyi Ji, Junjie Yang, Yingbin Liang

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We further validate our theoretical results and demonstrate the efficiency of bilevel optimization algorithms by the experiments on meta-learning and hyperparameter optimization. The experiments validate our theoretical results for deterministic bilevel optimization, and demonstrate the superior efficiency of stoc Bi O for stochastic bilevel optimization.
Researcher Affiliation Academia 1Department of Electrical and Computer Engineering, The Ohio State University.
Pseudocode Yes Algorithm 1 Bilevel algorithms via AID or ITD
Open Source Code Yes More details can be found in our code via https://github.com/Junjie Yang97/Stoc Bio_hp.
Open Datasets Yes We conduct experiments over a 5-way 5-shot task on two datasets: FC100 and mini Image Net.
Dataset Splits No The paper mentions using "training set", "held-out test sets", "validation and training data" but does not provide specific numerical split percentages or sample counts for the datasets (e.g., FC100, mini Image Net, 20 Newsgroup, MNIST) within the main text.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments.
Software Dependencies No Our implementation applies the function torch.autograd.grad in Py Torch, which automatically determines the size of Jacobians.
Experiment Setup Yes Due to the space limitations, we provide the model architectures and hyperparameter settings in Appendix A.