Research map / Continual learning / Online and replay-based
Continual Learning with Deep Generative Replay
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 |
|---|---|---|---|
| replay generated past inputs with targets from the previous solver | Deep Generative Replay · MNIST pixel-permutation tasks | average test accuracy on learned tasks | Helped |
| replay actual past inputs paired with previous solver predictions instead of generated inputs (ablation) | Deep Generative Replay · MNIST pixel-permutation tasks | average test accuracy on learned tasks | No clear effect |
| replace meaningful replay inputs with random Gaussian noise while retaining recorded responses (ablation) | Deep Generative Replay · MNIST pixel-permutation tasks | average test accuracy on learned tasks | Hurt |
| remove replay (ablation) | Deep Generative Replay solver · MNIST pixel-permutation tasks | average test accuracy on learned tasks | Hurt |
| replay generated past inputs | Deep Generative Replay · MNIST and SVHN, sequential domain learning | test accuracy on the first domain | Helped |
| replay actual past inputs paired with previous solver predictions instead of generated inputs (ablation) | Deep Generative Replay · MNIST and SVHN, sequential domain learning | test accuracy on the first domain | No clear effect |
| remove replay (ablation) | Deep Generative Replay solver · MNIST and SVHN, sequential domain learning | average test accuracy across domains | Hurt |
| add generated old-domain inputs to the LwF training procedure | LwF · SVHN then MNIST | test accuracy on SVHN | Helped |
| remove generative replay from LwF (ablation) | LwF-GR · SVHN then MNIST | test accuracy on SVHN | Hurt |
| replay random inputs rather than meaningful past inputs (ablation) | Deep Generative Replay · MNIST disjoint two-class tasks | test performance across encountered classes | Hurt |
| remove replay and train on the current class subset only (ablation) | Deep Generative Replay solver · MNIST disjoint two-class tasks | test performance across encountered classes | Hurt |
| reconstruct both cumulative input and target distributions through replay | Deep Generative Replay · MNIST disjoint two-class tasks | test performance across encountered classes | Helped |
Closest papers by meaning
- DDGR: Continual Learning with Deep Diffusion-based Generative Replay (ICML 2023 · similarity 0.92)
- Online Continual Learning with Maximal Interfered Retrieval (NeurIPS 2019 · similarity 0.90)
- Memory Replay GANs: Learning to Generate New Categories without Forgetting (NeurIPS 2018 · similarity 0.90)
- Task-Free Continual Generation and Representation Learning via Dynamic Expansionable Memory Cluster (AAAI 2024 · similarity 0.89)
- Recall-Oriented Continual Learning with Generative Adversarial Meta-Model (AAAI 2024 · similarity 0.89)
- Continual Learning in Modern Hopfield Networks with an Application to Diffusion Models (NeurIPS 2026 · similarity 0.88)
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.