Research map / Learning theory and algorithms / Prediction and decision losses
Curriculum Learning by Dynamic Instance Hardness
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 |
|---|---|---|---|
| DIH-guided curriculum (DIHCL-Exp, Loss) instead of random mini-batches | SGD · CIFAR10 | test accuracy: 96.18 -> 97.03 | Helped |
| DIH-guided curriculum (DIHCL-Exp, Loss) instead of random mini-batches | SGD · CIFAR100 | test accuracy: 79.64 -> 82.23 | Helped |
| DIH-guided curriculum (DIHCL-Exp, dLoss) instead of random mini-batches | SGD · STL10 | test accuracy: 86.06 -> 89.41 | Helped |
| DIH-guided curriculum (DIHCL-Beta, Flip) instead of random mini-batches | SGD · ImageNet | test accuracy: 75.04 -> 76.33 | Helped |
| DIH-guided curriculum instead of MCL (instantaneous loss) | MCL · Food-101 | test accuracy: 84.18 -> 84.94 | Helped |
| DIH-guided curriculum (DIHCL-Exp, dLoss) instead of MCL | MCL · Aircraft | test accuracy: 75.28 -> 79.72 | Helped |
| self-paced learning instead of random mini-batches | SGD · CIFAR10 | test accuracy: 96.18 -> 93.55 | Hurt |
| DIHCL-Exp dLoss instead of Rand Loss variant (instantaneous choice) | DIHCL · CIFAR10 | test accuracy: 96.76 -> 96.40 | Mixed |
| change discount factor gamma from 0.95 to 0.9 or 0.8 (ablation) | DIHCL-Rand Loss · CIFAR10 | test accuracy: 96.76 vs 96.75 vs 96.78 | No clear effect |
| variants of DIHCL: exponential decay LR, T0=3, increasing budget, smaller gamma_k, 40% noisy labels (ablation) | DIHCL · CIFAR10 | test accuracy | Helped |
Closest papers by meaning
- On The Power of Curriculum Learning in Training Deep Networks (ICML 2019 · similarity 0.88)
- Curriculum Learning by Transfer Learning: Theory and Experiments with Deep Networks (ICML 2018 · similarity 0.87)
- Robust Curriculum Learning: from clean label detection to noisy label self-correction (ICLR 2021 · similarity 0.86)
- Provable Advantage of Curriculum Learning on Parity Targets with Mixed Inputs (NeurIPS 2023 · similarity 0.86)
- Distilled Pretraining: A modern lens of Data, In-Context Learning and Test-Time Scaling (ICLR 2026 · similarity 0.85)
- Progressive distillation induces an implicit curriculum (ICLR 2025 · similarity 0.84)
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.