Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Periodic Activation Functions Induce Stationarity
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 |
|---|---|---|---|
| replace locally stationary Mat-3/2 activation with global (sin) Mat-3/2 | BNN last layer · UCI boston | NLPD: 0.71 -> 0.43 | Helped |
| replace ReLU with global Mat-3/2 (prelu) | BNN · UCI boston | NLPD: 0.51 -> 0.38 | Helped |
| replace ReLU with global Mat-3/2 (prelu) | BNN · UCI concrete | NLPD: 0.78 -> 0.72 | Helped |
| replace ReLU with global RBF (prelu) | BNN · UCI airfoil | NLPD: 0.51 -> 0.05 | Helped |
| replace ReLU with global RBF (prelu) | BNN · UCI airfoil | RMSE: 0.41 -> 0.26 | Helped |
| replace ReLU with global RBF (prelu) | BNN · UCI elevators | NLPD: 0.38 -> 0.74 | Hurt |
| replace ReLU with global Mat-3/2 (sin) | BNN · UCI elevators | NLPD: 0.38 -> 0.37 | No clear effect |
| replace ReLU with global Mat-3/2 (sin) | BNN · UCI concrete | RMSE: 0.48 -> 0.49 | No clear effect |
| replace locally stationary Mat-3/2 with global Mat-3/2 (sin) | BNN · UCI concrete | RMSE: 0.42 -> 0.49 | Hurt |
| use stationary (global) activations instead of ReLU | BNN · Rotated MNIST | NLPD / mean confidence | Helped |
| use stationary activations instead of ReLU, accuracy | BNN · Rotated MNIST | accuracy | No clear effect |
| use globally stationary Mat-3/2 sine activation instead of ReLU | BNN (SWAG) · CIFAR-10 vs CIFAR-100/SVHN OOD | predictive marginal variance | Mixed |
Closest papers by meaning
- Neural Functions for Learning Periodic Signal (ICLR 2025 · similarity 0.85)
- Linearization Turns Neural Operators into Function-Valued Gaussian Processes (ICML 2025 · similarity 0.83)
- FSP-Laplace: Function-Space Priors for the Laplace Approximation in Bayesian Deep Learning (NeurIPS 2024 · similarity 0.82)
- Global Convergence in Neural ODEs: Impact of Activation Functions (ICLR 2025 · similarity 0.82)
- Bayesian Neural Scaling Law Extrapolation with Prior-Data Fitted Networks (ICML 2025 · similarity 0.82)
- FINER: Flexible Spectral-bias Tuning in Implicit NEural Representation by Variable-periodic Activation Functions (CVPR 2024 · similarity 0.82)
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.