Research map / Continual learning / Domain and prompt-based
Life-Long Disentangled Representation Learning with Cross-Domain Latent Homologies
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 |
|---|---|---|---|
| Use VASE's cluster inference, dreaming feedback, and atypicality-based latent masking together (ablation) | Disentangled VASE · moving Fashion-MNIST → MNIST → moving MNIST | object ID accuracy, maximum (%): 86.7 ($\pm$1.9) -> 88.6 ($\pm$0.4) | Mixed |
| Add cluster inference to the disentangled method without cluster inference (ablation) | VASE (DA) · MNIST → Fashion-MNIST → MNIST | average decrease in classifier accuracy (%): -7.9 -> -0.9 | Helped |
| Add atypicality-based latent masking to the disentangled method without that mask (ablation) | VASE (SD) · MNIST → Fashion-MNIST → MNIST | average decrease in classifier accuracy (%): -2.2 -> -0.9 | Helped |
| Add dreaming feedback to the disentangled method without dreaming (ablation) | VASE (SA) · MNIST → Fashion-MNIST → MNIST | average decrease in classifier accuracy (%): -5.9 -> -0.9 | Helped |
| Use disentangled rather than entangled representations (ablation) | VASE · moving Fashion-MNIST → MNIST → moving MNIST | position MSE, minimum (*1e-4): 4.2 ($\pm$0.4) -> 3.2 ($\pm$0.03) | Mixed |
| Increase final capacity from 20 to 35 (ablation) | VASE · Hyperparameter search | classification accuracy: 20 to 35 | Helped |
| Lower the atypicality threshold (ablation) | VASE · Hyperparameter search | classification performance | No clear effect |
| Vary the capacity-constraint coefficient (ablation) | VASE · Hyperparameter search | classification performance | No clear effect |
| Vary the dreaming reference-network update frequency (ablation) | VASE · Hyperparameter search | classification performance | No clear effect |
| Use imagination-driven data augmentation during static MNIST training | VASE · MNIST training; moving MNIST classifier | classifier accuracy | Helped |
| Use disentangled rather than entangled representations for cross-domain inference | VASE · DMLab-30 NatLab ↔ EDE | semantic quality of cross-domain reconstructions | Helped |
Closest papers by meaning
- Detach and Adapt: Learning Cross-Domain Disentangled Deep Representation (CVPR 2018 · similarity 0.86)
- Orthogonality-Enforced Latent Space in Autoencoders: An Approach to Learning Disentangled Representations (ICML 2023 · similarity 0.85)
- Factorized Diffusion Autoencoder for Unsupervised Disentangled Representation Learning (AAAI 2024 · similarity 0.85)
- An Information Criterion for Controlled Disentanglement of Multimodal Data (ICLR 2025 · similarity 0.85)
- RePercENT: Scaling Disentangled Representation Learning Beyond Two Modalities (NeurIPS 2026 · similarity 0.85)
- Unity by Diversity: Improved Representation Learning for Multimodal VAEs (NeurIPS 2024 · similarity 0.85)
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.