Research map / Architectures and efficiency / RNNs and normalization
Learning Deep ℓ0 Encoders
Zhangyang Wang, Qing Ling, Thomas Huang
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 a two-stage learned Deep ℓ0-Regularized Encoder instead of 10 iterative updates | ℓ0 sparse code prediction · MNIST patches, ℓ0-regularized, p = 128, 256, 512 | prediction error (%): 3.55 -> 0.92; 4.33 -> 0.91; 4.08 -> 0.81 | Improved |
| Use a two-stage Deep M-Sparse Encoder instead of 10 iterative updates | M-sparse code prediction · MNIST patches, p = 128, 256, 512 | prediction error (%): 5.67 -> 2.94; 5.44 -> 2.87; 5.20 -> 3.29 | Improved |
| Use max_M pooling and unpooling to enforce the M-sparse support | Deep M-Sparse Encoder · MNIST patches, p = 128, 256, 512 | averaged non-zero support error: 4.6 -> 2.2; 5.6 -> 2.7; 5.3 -> 2.7 | Improved |
| Tune the enforced sparsity level to M = 20 rather than M = 10 or M = 30 (ablation) | Deep M-Sparse Encoder · MNIST classification, p = 128, 256, 512 | classification error rate (%): 0.72 -> 0.69; 0.57 -> 0.54; 0.53 -> 0.51 | Improved |
| Set the enforced sparsity level to M = 10 rather than M = 20 or M = 30 (ablation) | Deep M-Sparse Encoder · AVIRIS Indiana Pines classification, p = 128, 256, 512 | classification error rate (%): 14.67 -> 13.77; 14.23 -> 13.56; 14.07 -> 13.52 | Improved |
| Increase the number of unfolded stages from K = 2 to K = 3 or 4 (ablation) | Deep ℓ0 encoders · Several tested cases | performance | Improved |
| Use Deep M-Sparse (M = 10) rather than Deep ℓ0-Regularized Encoder (ablation) | Deep encoders · COIL 20 clustering, p = 512 | clustering error rate (%): 14.06 -> 14.03 | Improved |
| Use the learned-sparsity regularized encoder rather than hardcoding M (ablation) | Deep encoders · CMU PIE clustering | clustering error rate (%): 15.41 vs 16.01 | Improved |
| Use Deep ℓ0 encoders rather than Tuned LISTA | End-to-end sparse-code classification · MNIST and AVIRIS Indiana Pines | classification error rate (%): MNIST p = 512: 0.57 -> 0.51; AVIRIS p = 512: 15.97 -> 13.52 | Mixed |
Closest papers by meaning
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.88)
- A Single-Step, Sharpness-Aware Minimization is All You Need to Achieve Efficient and Accurate Sparse Training (NeurIPS 2024 · similarity 0.87)
- Global Minimizers of $\ell^p$-Regularized Objectives Yield the Sparsest ReLU Neural Networks (NeurIPS 2025 · similarity 0.86)
- Compute Optimal Inference and Provable Amortisation Gap in Sparse Autoencoders (ICML 2025 · similarity 0.86)
- Deep Weight Factorization: Sparse Learning Through the Lens of Artificial Symmetries (ICLR 2025 · similarity 0.86)
- Two Sparse Matrices are Better than One: Sparsifying Neural Networks with Double Sparse Factorization (ICLR 2025 · 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.
