Research map / Probabilistic and scientific modeling / Variational and generative sampling
On the relationship between variational inference and auto-associative memory
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 a latent representation and balanced-GMM retrieval rather than pixel-level GMM retrieval | GMM associative memory · CLEVR, shifted inputs | successful retrieval (%): 28±8.5 -> 63.1±10.2 | Helped |
| Train the precision matrix for the retrieval transformation | VAE-GMM · CLEVR color rotation | successful retrieval (%): 53.6±12.0 -> 98.9±2.5 | Helped |
| Train the precision matrix for the retrieval transformation | VAE-GMM · CLEVR shift | successful retrieval (%): 63.1±10.2 -> 93.7±5.3 | Helped |
| Add BP-based iterative inference to VAE-GMM | VAE-GMM · CIFAR10, noise σ=0.6 | successful retrieval (%): 86.3±9.1 vs 81.1±8.9 | No clear effect |
| Add BP-based iterative inference to VAE-GMM | VAE-GMM · CLEVR, noise σ=0.6 | successful retrieval (%): 4.4±4.4 vs 2.0±2.5 | No clear effect |
| Use the VAE latent representation rather than pixel-space memory matching | GMM associative memory · CLEVR, noise σ=0.6 | successful retrieval (%): 73.0±10.1 -> 4.4±4.4 | Hurt |
| Remove VAE encoder initialization (ablation) | BP-GMM · CLEVR corrupted inputs | successful retrieval (%) | Hurt |
| Remove VAE encoder initialization (ablation) | BP-GMM · CIFAR10 corrupted inputs | successful retrieval (%) | No clear effect |
| Replace the balanced GMM prior with the biased Hopfield prior (ablation) | VAE-BP-GMM · CIFAR10 corrupted inputs | successful retrieval (%) | Mixed |
| Replace pixel-level addressing with latent-representation addressing | Associative memory models · CLEVR color rotation and shift | successful retrieval (%): 48±16.9 -> 53.6±12.0; 0.7±1.9 -> 63.1±10.2 | Helped |
| Use the balanced GMM prior instead of the biased Hopfield prior (ablation) | Associative memory · CIFAR10 clean inputs | successful retrieval (%): 3.0±3.9 -> 100±0.0 | Helped |
Closest papers by meaning
- Inference Suboptimality in Variational Autoencoders (ICML 2018 · similarity 0.89)
- Variational Memory Addressing in Generative Models (NeurIPS 2017 · similarity 0.88)
- Brain-like Variational Inference (NeurIPS 2025 · similarity 0.87)
- Factual recall in linear associative memories: sharp asymptotics and mechanistic insights (NeurIPS 2026 · similarity 0.87)
- Semantically-correlated memories in a dense associative model (ICML 2024 · similarity 0.86)
- Task-Free Continual Generation and Representation Learning via Dynamic Expansionable Memory Cluster (AAAI 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.