Research map / Continual learning / Online and replay-based
Lifelong Generative Modelling Using Dynamic Expansion Graph Model
Fei Ye, Adrian G. Bors
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 |
|---|---|---|---|
| Use DEGM-IWELBO-50 instead of the strongest listed GR baseline | Lifelong VAE · Split MNIST | log-likelihood: -93.57 -> -88.04 | Improved |
| Increase DEGM IWELBO weighted samples from 5 to 50 (ablation) | DEGM · Split MNIST | log-likelihood: -91.44 -> -88.04 | Improved |
| Use DEGM-IWELBO-50 instead of the strongest listed GR baseline | Lifelong VAE · Split Fashion | log-likelihood: -236.66 -> -233.76 | Improved |
| Increase DEGM IWELBO weighted samples from 5 to 50 (ablation) | DEGM · Split Fashion | log-likelihood: -235.93 -> -233.76 | Improved |
| Use DEGM-IWELBO-50 instead of the strongest listed GR baseline | Lifelong VAE · COFMI | log-likelihood: -172.10 -> -163.27 | Improved |
| Increase DEGM IWELBO weighted samples from 5 to 50 (ablation) | DEGM · COFMI | log-likelihood: -164.99 -> -163.27 | Improved |
| Use dynamic expansion and graph-based transfer rather than one VAE per task | Lifelong VAE · CCCOSCZC average | square loss: 381.3 -> 393.8 | Hurt |
| Use DEGM rather than one VAE per task | Lifelong VAE · CCCOSCZC | square loss: CelebA: 217.0 -> 229.2; Zappos: 275.8 -> 263.4 | Mixed |
| Add a second stochastic layer (ablation) | GR VAE · Split MNIST and Split Fashion | log-likelihood: Split MNIST: -93.57 -> -91.23; Split Fashion: -236.66 -> -236.90 | Mixed |
| Add a second stochastic layer (ablation) | GR VAE · COFMI | log-likelihood: -172.10 -> -188.9 | Hurt |
| Use IWELBO with 50 samples rather than ELBO (ablation) | DEGM · COFMI | log-likelihood: -168.91 -> -163.27 | Improved |
| Use IWELBO with 50 samples rather than ELBO (ablation) | DEGM · Split MNIST and Split Fashion | log-likelihood: Split MNIST: -93.51 -> -88.04; Split Fashion: -238.54 -> -233.76 | Improved |
Closest papers by meaning
- Continual Unsupervised Generative Modeling (TPAMI 2025 · similarity 0.94)
- Lifelong Variational Autoencoder via Online Adversarial Expansion Strategy (AAAI 2023 · similarity 0.91)
- Task-Free Continual Generation and Representation Learning via Dynamic Expansionable Memory Cluster (AAAI 2024 · similarity 0.89)
- Dynamic Expansion Diffusion Learning for Lifelong Generative Modelling (AAAI 2025 · similarity 0.88)
- Knowledge Graph Enhanced Generative Multi-modal Models for Class-Incremental Learning (NeurIPS 2025 · similarity 0.87)
- Recall-Oriented Continual Learning with Generative Adversarial Meta-Model (AAAI 2024 · 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.
