AI Paper Atlas

Research map / Continual learning / Online and replay-based

BooVAE: Boosting Approach for Continual Learning of VAE

Evgenii Egorov, Anna Kuzina, Evgeny Burnaev

NeurIPS 2021 · Continual learning › Online and replay-based · approach: Unified and unsupervised continual frameworks · 14 citations (OpenAlex)

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 changedWhereMeasuredEffect
learn prior components as encoder-evaluated pseudo-inputs (BooVAE) instead of latent-space Gaussian mixture (MoG) (ablation)VAE continual learning · MNIST (10 tasks)test NLL: 180.1 (1.3) -> 124.5 (1.9)Improved
learned pseudo-input prior instead of random coreset (ablation)VAE continual learning · MNIST (10 tasks)diversity KL: 2.09 (0.01) -> 0.11 (0.15)Improved
learned pseudo-input prior instead of random coreset (ablation)VAE continual learning · notMNIST (10 tasks)test NLL: 233.7 (2.3) -> 261.4 (8.7)Hurt
combine BooVAE with VCL (ablation)BooVAE · MNIST (10 tasks)test NLL: 124.5 (1.9) -> 112.7 (1.1)Improved
combine BooVAE with VCL (ablation)BooVAE · MNIST (10 tasks)diversity KL: 0.11 (0.15) -> 0.74 (0.01)Hurt
add EWC to the multihead VAE (ablation)Multihead · notMNIST (10 tasks)test NLL: 951.2 (47) -> 158.7 (2.2)Improved
use shared single-architecture BooVAE instead of multiheadVAE · CelebA hair tasksFID: 27.5 (task 1 BooVAE)Improved
BooVAE vs standard VAE without continual-learning regularizationVAE · MNIST (10 tasks)test NLL: 186.8 (2.3) -> 124.5 (1.9)Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy