Research map / Learning theory and algorithms / Prediction and decision losses
On the Pitfalls of Instance-Based Dynamic Curricula
Alexandre Galashov, Amal Rannen-Triki, Yee Whye Teh, Razvan Pascanu, András György
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
The paper examines instance-based dynamic curricula for neural network training, questioning whether recent reported gains from such curricula on image classification hold up.
Problem
Instance-based dynamic curricula (automatically ordering or weighting training examples) have been reported to speed up neural network training on image classification, but the conditions under which they help are unclear.
Approach
Critically analyzes prior instance-based dynamic curriculum methods (Jiang et al. 2019; Zhou et al. 2020; Wang et al. 2024b; Yuan et al. 2025) and highlights pitfalls of these approaches; the abstract only partially states the findings.
Experimental setup
Builds on: curriculum learning
Closest papers by meaning
- Learning to Learn: How to Continuously Teach Humans and Machines (ICCV 2023 · similarity 0.87)
- When Do Curricula Work? (ICLR 2021 · similarity 0.86)
- On The Power of Curriculum Learning in Training Deep Networks (ICML 2019 · similarity 0.86)
- Curriculum Coarse-to-Fine Selection for High-IPC Dataset Distillation (CVPR 2025 · similarity 0.83)
- When Dynamic Data Selection Meets Data Augmentation: Achieving Enhanced Training Acceleration (ICML 2025 · similarity 0.83)
- CurBench: Curriculum Learning Benchmark (ICML 2024 · 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.
