Research map / Probabilistic and scientific modeling / Molecules and proteins
Learning Overcomplete, Low Coherence Dictionaries with Linear Inference
Jesse A. Livezey, Alejandro F. Bujan, Friedrich T. Sommer
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 |
|---|---|---|---|
| replace L2 coherence cost with L4 cost | overcomplete ICA · 2x overcomplete dictionary, 32-dim, numerical minimization | pairwise angle distribution / coherence | Improved |
| use L2 cost from noisy pathological initialization | overcomplete ICA coherence control · 2x overcomplete, 32-dim | coherence | Hurt |
| remove quadratic Taylor term from Coulomb and Random Prior costs (flattened versions) (ablation) | Coulomb and Random Prior costs · 2x overcomplete, 32-dim, 10 random inits | coherence / pairwise angles near 90 degrees | Improved |
| use Coulomb or Random Prior cost instead of L2 at overcompleteness above 2 | coherence control cost · 32-dim, varying overcompleteness | median minimum pairwise angle | Improved |
| use L2 cost or Score Matching instead of other ICA costs | overcomplete ICA mixing matrix recovery · 32-dim, k=12 sparse, 2x overcomplete | normalized recovery error | Hurt |
| use sparse coding instead of ICA with linear inference at high overcompleteness | mixing matrix recovery · 32-dim, k=12, large overcompleteness | normalized recovery error | Improved |
| use L2 cost on natural image patches | 2x overcomplete ICA · Van Hateren 8x8 whitened patches | duplicate bases / coherence | Hurt |
Closest papers by meaning
- On the Identifiability of Nonlinear ICA: Sparsity and Beyond (NeurIPS 2022 · similarity 0.87)
- SPLICE: Fully Tractable Hierarchical Extension of ICA with Pooling (ICML 2017 · similarity 0.86)
- Robustness of Nonlinear Representation Learning (ICML 2024 · similarity 0.84)
- Feature learning from non-Gaussian inputs: the case of Independent Component Analysis in high dimensions (ICML 2025 · similarity 0.84)
- Cross-Entropy Is All You Need To Invert the Data Generating Process (ICLR 2025 · similarity 0.84)
- Diverse Influence Component Analysis: A Geometric Approach to Nonlinear Mixture Identifiability (NeurIPS 2025 · similarity 0.84)
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.
