Research map / Continual learning / Class-incremental
Learning to Learn: How to Continuously Teach Humans and Machines
Parantak Singh, You Li, Ankur Sikarwar, Stan Weixian Lei, Difei Gao, Morgan B. Talbot, Ying Sun, Mike Zheng Shou, Gabriel Kreiman, Mengmi 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 |
|---|---|---|---|
| choose a different class order (curriculum) in online class-incremental learning | Vanilla CL · MNIST/FashionMNIST/CIFAR10 | average accuracy alpha: alpha ranged from 19% to 26% depending on curriculum | Improved |
| use CD-ranked curricula instead of random ranking | Curriculum Designer · MNIST and CIFAR10 (Recall@K) | Recall@K: peaks of 0.5, 0.2, 1 at K=2, 5, 10 for MNIST, FashionMNIST, CIFAR10 | Improved |
| use CD instead of random designer | Curriculum Designer · FashionMNIST | Recall@K | Hurt |
| use Euclidean instead of cosine distance for class prototypes (ablation) | Curriculum Designer · MNIST | Recall@K | No clear effect |
| use OTDD instead of cosine distance (ablation) | Curriculum Designer · MNIST | Recall@K | Hurt |
| use earlier layers (6 or 11) rather than layer 12 features (ablation) | Curriculum Designer · MNIST | Recall@K | Hurt |
| replace SqueezeNet with ResNet34/ResNet18 teacher (ablation) | Curriculum Designer · MNIST | Recall@K | Mixed |
| pre-train the teacher on MNIST instead of ImageNet-100 (ablation) | Curriculum Designer · MNIST | Recall@K | No clear effect |
| use a pre-trained teacher rather than randomly initialized teacher (ablation) | Curriculum Designer · MNIST | Recall@K | Improved |
| compare CL algorithm rankings between algorithms versus against random designer | curricula ranking agreement · MNIST (paradigm-I) | Spearman correlation: 0.0002 -> 0.26 (algorithm-random vs between-algorithms) | Improved |
| use curricula ranked by algorithms for human learners versus random | NOD human experiment · NOD | curriculum discrepancy H: H decrease of 0.13 from random-humans to algorithm-humans | Improved |
| apply CD curricula to humans | Curriculum Designer · NOD | curriculum discrepancy H | No clear effect |
Closest papers by meaning
- A Survey on Curriculum Learning (TPAMI 2021 · similarity 0.91)
- Curriculum Learning: A Survey (IJCV 2022 · similarity 0.88)
- An Analytical Theory of Curriculum Learning in Teacher-Student Networks (NeurIPS 2022 · similarity 0.88)
- CurBench: Curriculum Learning Benchmark (ICML 2024 · similarity 0.87)
- Class Incremental Learning with Multi-Teacher Distillation (CVPR 2024 · similarity 0.86)
- Optimal Task Order for Continual Learning of Multiple Tasks (ICML 2025 · similarity 0.85)
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.
