Research map / Learning theory and algorithms / Prediction and decision losses
Bilevel Coreset Selection in Continual Learning: A New Formulation and Algorithm
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 |
|---|---|---|---|
| add bilevel optimizer for coreset weights (vs random init, i.e. Uniform) (ablation) | BCSR · Split CIFAR-100 | AVG ACC: 58.37 -> 61.60 | Helped |
| add bilevel optimizer for coreset weights (ablation) | BCSR · Permuted MNIST | AVG ACC: 53.34 -> 56.23 | Helped |
| add bilevel optimizer for coreset weights (ablation) | BCSR · Multiple Datasets | AVG ACC: 55.50 -> 59.89 | Helped |
| add bilevel optimizer for coreset weights (ablation) | BCSR · Split CIFAR-100 | FGT: 0.073 -> 0.051 | Helped |
| add smooth top-K regularizer (lambda 0 -> 0.10) (ablation) | BCSR · Split CIFAR-100 | AVG ACC: 60.43 -> 61.60 | Helped |
| increase lambda too far (0.10 -> 1.00) (ablation) | BCSR · Split CIFAR-100 | AVG ACC: 61.60 -> 58.89 | Hurt |
| larger lambda on Multiple Datasets (0.00 -> 1.00) (ablation) | BCSR · Multiple Datasets | AVG ACC: 55.71 -> 58.18 | Helped |
| increase coreset size K from 10 to 40 (ablation) | BCSR · Split CIFAR-100 | AVG ACC: 61.60 -> 56.58 | Hurt |
| increase inner loops N from 1 to 20 (ablation) | BCSR · Split CIFAR-100 | AVG ACC: 61.60 -> 59.20 | Hurt |
| increase inner loops N from 1 to 5 (ablation) | BCSR · Split CIFAR-100 | AVG ACC: 61.60 -> 61.75 | No clear effect |
| reduce Hessian-inverse loops Q from 3 to 1 (ablation) | BCSR · Split CIFAR-100 | AVG ACC: 61.60 -> 52.14 | Hurt |
| increase Q from 3 to 20 (ablation) | BCSR · Split CIFAR-100 | AVG ACC: 61.60 -> 57.80 | Hurt |
Closest papers by meaning
- Stable Coresets via Posterior Sampling: Aligning Induced and Full Loss Landscapes (NeurIPS 2025 · similarity 0.85)
- Bilevel Optimization under Unbounded Smoothness: A New Algorithm and Convergence Analysis (ICLR 2024 · similarity 0.84)
- A Nearly Optimal Single Loop Algorithm for Stochastic Bilevel Optimization under Unbounded Smoothness (ICML 2024 · similarity 0.84)
- Refined Coreset Selection: Towards Minimal Coreset Size under Model Performance Constraints (ICML 2024 · similarity 0.84)
- Pareto Continual Learning: Preference-Conditioned Learning and Adaption for Dynamic Stability-Plasticity Trade-off (AAAI 2025 · similarity 0.84)
- Learning Dynamics in Continual Pre-Training for Large Language Models (ICML 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.