Research map / Probabilistic and scientific modeling / Variational and generative sampling
Dense associative memory for Gaussian distributions
Chandan Tankala, Krishna Balasubramanian
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 Bures-Wasserstein geometry (BW-DAM) instead of Euclidean DAM on vectorized mean and covariance (ablation) | dense associative memory for Gaussians · CelebA masked-image retrieval (VAE latent Gaussians) | retrieval accuracy | Improved |
| use BW-DAM instead of pixel-space DAM (ablation) | dense associative memory for Gaussians · CelebA masked-image retrieval (VAE latent Gaussians) | retrieval accuracy | Improved |
| use BW-DAM instead of Euclidean DAM and Pixel-DAM (ablation) | dense associative memory for Gaussians · CIFAR-10 masked-image retrieval | retrieval accuracy | Improved |
| use BW-DAM instead of Euclidean DAM on five well-separated 2D Gaussians (ablation) | dense associative memory for Gaussians · synthetic 5 Gaussians in R^2, queries at 0.5r | retrieval success: Eu-DAM fails on two of five retrievals; BW-DAM retrieves all five | Improved |
| increase inverse temperature beta (ablation) | BW-DAM · text8 Word2Gauss embeddings (11,815 words, 50-d) | retrieval accuracy: low for small beta -> approaches 100% for large beta | Improved |
| raise beta from 0.1 to 2 (ablation) | BW-DAM · synthetic Gaussians on Wasserstein sphere, d=10 and d=20 | retrieval W2 distance: beta=0.1 fails entirely; beta=2 converges in a single iteration | Improved |
| raise beta on high-beta word retrieval (separation condition possibly violated) (ablation) | BW-DAM · text8 Word2Gauss embeddings | retrieval accuracy | Mixed |
Closest papers by meaning
- Dense Associative Memory with Epanechnikov Energy (NeurIPS 2025 · similarity 0.90)
- Uniform Memory Retrieval with Larger Capacity for Modern Hopfield Models (ICML 2024 · similarity 0.86)
- Learning Associative Memories with Gradient Descent (ICML 2024 · similarity 0.86)
- Provably Optimal Memory Capacity for Modern Hopfield Models: Transformer-Compatible Dense Associative Memories as Spherical Codes (NeurIPS 2024 · similarity 0.86)
- Factual recall in linear associative memories: sharp asymptotics and mechanistic insights (NeurIPS 2026 · similarity 0.86)
- Dense Associative Memory Through the Lens of Random Features (NeurIPS 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.
