Research map / Continual learning / Online and replay-based
Simple Lifelong Learning Machines
Joshua T. Vogelstein, Jayanta K. Dey, Hayden S. Helm, Will LeVine, Ronak Mehta, Tyler M. Tomita, Haoyin Xu, Ali Geisa, Qingyang Wang, Gido M. van de Ven, Chenyu Gao, Weiwei Yang and 4 more
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 |
|---|---|---|---|
| replay more old task data (40% to 100%) (ablation) | SiLLy-N · CIFAR 10X10 | backward transfer | Improved |
| replay more old task data (40% to 100%) (ablation) | SiLLy-N · CIFAR 10X10 | forward transfer | No clear effect |
| allow resource growth instead of constant 4 encoders (ablation) | SiLLy-N · CIFAR 10X10 | accuracy | Improved |
| constant resource mode with 4 encoders instead of growing (ablation) | SiLLy-N-4 · CIFAR 10X10 | forward and backward transfer | No clear effect |
| use pretrained encoders (ResNet 50, ViT_B16) (ablation) | SiLLy-N · CIFAR 10X10 | forward transfer and accuracy | Mixed |
| use ViT encoder instead of ResNet 50 (ablation) | SiLLy-N · CIFAR 10X10 | accuracy | Hurt |
| shuffle class labels in tasks 2 to 10 (adversarial) | SiLLy-N · CIFAR 10X10 | backward transfer | No clear effect |
| permute task order | SiLLy-N · CIFAR 10X10 | Transfer | No clear effect |
| use more than 30 encoders in constant resource budget (ablation) | SiLLy-N · FOOD1k 50X20 | transfer | No clear effect |
| use random forest encoders (SiLLy-F) on flattened tabular CIFAR 10X10 | SiLLy-F · tabular CIFAR 10X10 | transfer | Improved |
| use boosting (Model Zoo) instead of bagging (SiLLy-N) with many samples per task | SiLLy-N · Split Mini-Imagenet and 5-dataset | transfer | Hurt |
| rotate second subset away from 45 degrees toward 90 degrees (ablation) | SiLLy-N · Rotated XOR | backward transfer: increases back up to 0.18 at 90 degrees | Improved |
Closest papers by meaning
- Is Forgetting Less a Good Inductive Bias for Forward Transfer? (ICLR 2023 · similarity 0.90)
- Probing Representation Forgetting in Supervised and Unsupervised Continual Learning (CVPR 2022 · similarity 0.88)
- Encoder Based Lifelong Learning (ICCV 2017 · similarity 0.88)
- Do Your Best and Get Enough Rest for Continual Learning (CVPR 2025 · similarity 0.88)
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.87)
- Adaptive Retention & Correction: Test-Time Training for Continual Learning (ICLR 2025 · similarity 0.87)
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.
