Research map / Probabilistic and scientific modeling / Variational and generative sampling
Compressed Sensing using Generative Models
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 pretrained generator prior instead of pixel-space Lasso | MNIST compressed sensing · MNIST test set, Gaussian measurements | reconstruction error: 25 measurements; Lasso uses 400 measurements for matching performance | Helped |
| Use a pretrained generator prior instead of DCT or wavelet Lasso | CelebA compressed sensing · CelebA test set, Gaussian measurements | reconstruction quality: 500 measurements | Helped |
| Increase measurements beyond the generator's effective range | MNIST compressed sensing · MNIST test set, more than 500 measurements | reconstruction error: 100 measurements; more than 500 measurements | Mixed |
| Increase measurements beyond the generator's effective range | CelebA compressed sensing · CelebA test set, more than 5000 measurements | reconstruction error: 5000 measurements | Mixed |
| Add the latent prior regularizer λ‖z‖² (ablation) | Generative compressed sensing · MNIST and CelebA reconstruction | reconstruction error | Helped |
| Use fixed measurement matrix rather than a different random matrix for every image (ablation) | End-to-end MNIST reconstruction · MNIST, 10, 20, and 30 measurements | reconstruction error: 10, 20 and 30 measurements | Helped |
| Optimize the measurement matrix jointly rather than fixing it at random (ablation) | End-to-end MNIST reconstruction · MNIST, 10, 20, and 30 measurements | reconstruction error: 10, 20 and 30 measurements | Mixed |
| Use a larger, more flexible generator | Generative compressed sensing · MNIST and CelebA reconstruction | reconstruction error | Helped |
| Use generative compressed sensing rather than Lasso as measurement noise increases | MNIST and CelebA reconstruction · MNIST and CelebA, noise-tolerance experiment | reconstruction error: m=500 for MNIST; m=2500 for celebA | Helped |
Closest papers by meaning
- On the Power of Compressed Sensing with Generative Models (ICML 2020 · similarity 0.96)
- Modeling Sparse Deviations for Compressed Sensing using Generative Models (ICML 2018 · similarity 0.94)
- Sample Complexity Bounds for 1-bit Compressive Sensing and Binary Stable Embeddings with Generative Priors (ICML 2020 · similarity 0.93)
- Sparse Bayesian Generative Modeling for Compressive Sensing (NeurIPS 2024 · similarity 0.89)
- A Unified Framework for Learning with Nonlinear Model Classes from Arbitrary Linear Samples (ICML 2024 · similarity 0.88)
- The Price of Sparsity: Sufficient Conditions for Sparse Recovery using Sparse and Sparsified Measurements (NeurIPS 2025 · similarity 0.87)
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.