Research map / Continual learning / Online and replay-based
Memory Replay GANs: Learning to Generate New Categories without Forgetting
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 |
|---|---|---|---|
| Add conditional-generator memory replay and train jointly with replay samples | Sequential conditional GAN · MNIST, 5 tasks (0-4) | classification accuracy (%): 90.39 -> 97.93 | Helped |
| Add conditional-generator memory replay and train jointly with replay samples | Sequential conditional GAN · SVHN, 5 tasks (0-4) | classification accuracy (%): 61.29 -> 80.90 | Helped |
| Add replay alignment using pixelwise L2 loss against the previous generator | Sequential conditional GAN · LSUN, after 4th task | FID: 93.70 -> 37.73 | Helped |
| Use replay alignment instead of joint training with replay | MeRGAN · LSUN, after 4th task | FID: 49.69 -> 37.73 | Helped |
| Use replay alignment instead of joint training with replay | MeRGAN · SVHN, 5 tasks (0-4) | classification accuracy (%): 80.90 -> 76.05 | Hurt |
| Replace AC-GAN with cGAN using a projection discriminator and one-hot conditioning, and use ResNet-18 | MeRGAN on SVHN · SVHN, 10 tasks (0-9) | reverse classification accuracy: 20.07 -> 51.00; 49.43 -> 70.98 | Helped |
Closest papers by meaning
- Continual Learning with Deep Generative Replay (NeurIPS 2017 · similarity 0.90)
- Memorization Precedes Generation: Learning Unsupervised GANs with Memory Networks (ICLR 2018 · similarity 0.90)
- Recall-Oriented Continual Learning with Generative Adversarial Meta-Model (AAAI 2024 · similarity 0.87)
- Continual Learning in Modern Hopfield Networks with an Application to Diffusion Models (NeurIPS 2026 · similarity 0.86)
- Understanding the Forgetting of (Replay-based) Continual Learning via Feature Learning: Angle Matters (ICML 2025 · similarity 0.86)
- Predicting the Susceptibility of Examples to Catastrophic Forgetting (ICML 2025 · similarity 0.86)
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.