Research map / Learning theory and algorithms / Matrix and tensor methods
Unrolling PALM for Sparse Semi-Blind Source Separation
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 |
|---|---|---|---|
| Unroll PALM and learn its parameters using factor-supervised training | Sparse semi-blind source separation · Synthetic astrophysical test set | Median NMSE (sources): 0.0176 -> 0.00085 | Helped |
| Replace iterative PALM with a 25-layer LPALM network | Sparse semi-blind source separation · Synthetic astrophysical test set | Iterations/layers required: 2011 iterations -> 25 layers | Helped |
| Use LPALM instead of PALM with random initializations | Sparse semi-blind source separation · One synthetic astrophysical test sample | Median NMSE (sources): 0.0201 -> 0.0010 | Helped |
| Train across samples with variable mixing matrices rather than on one matrix | Realistic astrophysical source separation · Realistic test set, model evaluation | NMSE (mixing matrix A): 0.0088 -> 0.0009 | Helped |
| Use LISTA-CP source updates instead of ISTA-LLT source updates (ablation) | Unrolled PALM · Synthetic astrophysical test set | Median NMSE (sources): 0.7943 -> 0.00085 | Helped |
| Replace learned A step sizes with an explicit A update (ablation) | LPALM with LISTA-CP S updates · Synthetic astrophysical test set | Median NMSE (sources): 0.00085 -> 0.00097 | No clear effect |
| Use an explicit A update instead of learned A step sizes, alongside ISTA-LLT S updates (ablation) | Unrolled PALM · Synthetic astrophysical test set | Median NMSE (sources): 0.7943 -> 0.7948 | No clear effect |
| Replace factor supervision with unsupervised reconstruction loss (ablation) | LPALM · Synthetic astrophysical test set | Median NMSE (sources) + NMSE (mixing matrix A): 0.00112 -> 0.568 | Hurt |
| Supervise only the source estimates rather than both factors (ablation) | LPALM · Synthetic astrophysical test set | Median NMSE (sources) + NMSE (mixing matrix A): 0.00112 -> 0.00115 | No clear effect |
| Supervise only the mixing matrix estimates rather than both factors (ablation) | LPALM · Synthetic astrophysical test set | Median NMSE (sources) + NMSE (mixing matrix A): 0.00112 -> 0.00191 | No clear effect |
| Increase the number of LPALM layers beyond the best setting (ablation) | LPALM · Synthetic astrophysical validation data | Sum of NMSE for sources and mixing matrix | Hurt |
| Use noisy mixing matrix estimates in LISTA-CP updates (ablation) | LISTA-CP source estimator · Variable-mixing-matrix experiment | NMSE (sources) | Hurt |
Closest papers by meaning
- A Data-Driven Prism: Multi-View Source Separation with Diffusion Model Priors (NeurIPS 2025 · similarity 0.82)
- Unmixing Before Fusion: A Generalized Paradigm for Multi-Source-based Hyperspectral Image Synthesis (CVPR 2024 · similarity 0.81)
- Proximal Algorithm Unrolling: Flexible and Efficient Reconstruction Networks for Single-Pixel Imaging (CVPR 2025 · similarity 0.81)
- Unmixing Diffusion for Self-Supervised Hyperspectral Image Denoising (CVPR 2024 · similarity 0.81)
- SSDiff: Spatial-spectral Integrated Diffusion Model for Remote Sensing Pansharpening (NeurIPS 2024 · similarity 0.80)
- Untrained Neural Nets for Snapshot Compressive Imaging: Theory and Algorithms (NeurIPS 2024 · similarity 0.80)
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.