Research map / Architectures and efficiency / RNNs and normalization
Deep Learning with Learnable Product-Structured Activations
Saanjali Maharaj, Prasanth B. Nair
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 |
|---|---|---|---|
| add product-structured learnable activations (LRNN) to SPDER | SPDER · Cameraman image (~197k params) | PSNR (dB): 49.0 -> 107.9 | Improved |
| add product-structured learnable activations (LRNN) | SPDER · Audio representation: bach | MSE loss (x1e-4): 1.12 -> 0.10 | Improved |
| add product-structured learnable activations (LRNN) | SPDER · Audio representation: reggae | MSE loss (x1e-4): 24.8 -> 7.93 | Improved |
| add product-structured learnable activations (LRNN) | SPDER · Audio representation: reading | MSE loss (x1e-4): 8.88 -> 1.86 | Improved |
| add product-structured learnable activations (LRNN) | SPDER · Audio representation: reading | frequency similarity rho_AG: 0.9324 -> 0.9862 | Improved |
| use LRNN-SPDER instead of SPDER/SIREN for 40 dB target | SIREN, SPDER · ImageNet 1000 images, 40 dB target | success rate (%): 100% vs failures of 98.2% (SIREN) and 73.6% (SPDER) | Improved |
| use LRNN instead of WIRE | WIRE · Sparse-view CT 256x256 chest | PSNR (dB): 28.83 -> 29.13 | Improved |
| use LRNN instead of WIRE | WIRE · Sparse-view CT 256x256 chest | SSIM: 0.6413 -> 0.7455 | Improved |
| use 16k-parameter LRNN instead of 132k-parameter SIREN | SIREN · High-frequency Poisson PDE (n=2) | MSE: 132k -> 16k parameters | Improved |
| add LayerNorm after each LRNN layer (ablation) | LRNN · deep LRNN training | convergence | Improved |
| use periodic activations in univariate components (ablation) | LRNN · high-frequency signal representation | spectral bias | Improved |
| share activation functions across neurons in each layer (ablation) | LRNN · image representation | fidelity vs parameter count | Mixed |
Closest papers by meaning
- Maestro: Uncovering Low-Rank Structures via Trainable Decomposition (ICML 2024 · similarity 0.86)
- Learning Reveals Invisible Structure in Low-Rank RNNs (NeurIPS 2026 · similarity 0.85)
- Efficient Learning with Sine-Activated Low-Rank Matrices (ICLR 2025 · similarity 0.85)
- Position: A Theory of Deep Learning Must Include Compositional Sparsity (ICML 2025 · similarity 0.85)
- Global Convergence and Rich Feature Learning in $L$-Layer Infinite-Width Neural Networks under $\mu$ Parametrization (ICML 2025 · similarity 0.84)
- Better NTK Conditioning: A Free Lunch from (ReLU) Nonlinear Activation in Wide Neural Networks (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.
