Research map / Continual learning / Online and replay-based
Dream2Learn: Structured Generative Dreaming for Continual Learning
Salvatore Calcagno, Matteo Pennisi, Federica Proietto Salanitri, Amelia Sorrenti, Simone Palazzo, Concetto Spampinato, Giovanni Bellitto
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 |
|---|---|---|---|
| add D2L dream classes | ER-ACE · Mini-ImageNet class-incremental (buffer 2000) | FAA: 33.26 -> 40.90 | Improved |
| add D2L dream classes | ER · Mini-ImageNet class-incremental (buffer 5000) | FAA: 34.21 -> 39.75 | Improved |
| add D2L dream classes | DER++ · Mini-ImageNet class-incremental (buffer 2000) | FAA: 14.74 -> 21.06 | Improved |
| add D2L dream classes | ER-ACE · ImageNet-R class-incremental (buffer 2000) | FAA: 7.09 -> 9.54 | Improved |
| add D2L dream classes to improve forward transfer | ER · Mini-ImageNet (buffer 2000) | forward transfer: -2.58 -> +0.33 | Improved |
| add D2L dream classes to improve forward transfer | ER-ACE · Mini-ImageNet (buffer 5000) | forward transfer: -2.45 -> -1.58 | Hurt |
| add oracle-based stopping instead of fixed optimization (ablation) | D2L · Mini-ImageNet class-incremental (buffer 2000) | FAA: 38.94 -> 40.90 | Improved |
| add oracle-based stopping instead of fixed optimization (ablation) | D2L · Mini-ImageNet class-incremental (buffer 5000) | FAA: 46.41 -> 47.32 | Improved |
| replace dream classes each task instead of generating only at beginning (ablation) | D2L · Mini-ImageNet class-incremental (buffer 2000) | FAA: 36.93 -> 40.90 | Improved |
| replace dream classes each task instead of accumulating them (ablation) | D2L · Mini-ImageNet class-incremental (buffer 5000) | FAA: 43.87 -> 47.32 | Improved |
| use dream classes instead of Mixup (ablation) | D2L · Mini-ImageNet class-incremental (buffer 2000) | FAA: 36.84 -> 40.90 | Improved |
| use dream classes instead of Textual Mixup (ablation) | D2L · Mini-ImageNet class-incremental (buffer 2000) | FAA: 31.89 -> 40.90 | Improved |
Closest papers by meaning
- Task-Free Continual Generation and Representation Learning via Dynamic Expansionable Memory Cluster (AAAI 2024 · similarity 0.86)
- Recall-Oriented Continual Learning with Generative Adversarial Meta-Model (AAAI 2024 · similarity 0.85)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.85)
- Synthetic Data is an Elegant GIFT for Continual Vision-Language Models (CVPR 2025 · similarity 0.84)
- Stabilizing Zero-Shot Prediction: A Novel Antidote to Forgetting in Continual Vision-Language Tasks (NeurIPS 2024 · similarity 0.84)
- Predictive Attractor Models (NeurIPS 2024 · 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.
