Research map / Probabilistic and scientific modeling / Variational and generative sampling
A Probabilistic Framework for Nonlinearities in Stochastic Neural Networks
Qinliang Su, xuejun Liao, Lawrence Carin
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 |
|---|---|---|---|
| Learn unit-specific truncation points instead of using the original RBM nonlinearity | RBM · MNIST | average test log-probability: -86.3 -> -82.5 | Improved |
| Learn unit-specific truncation points instead of using the original RBM nonlinearity | RBM · Caltech101 Silhouettes | average test log-probability: -109.0 -> -104.3 | Improved |
| Use ReLU-type truncation points [0, +∞) rather than sigmoid-type [0, 1] or tanh-type [-1, 1] (ablation) | TruG-RBM · MNIST | average test log-probability: -97.3; -83.2; -124.5 | Mixed |
| Use unit-specific learned nonlinearities rather than fixed ReLU-type truncation points (ablation) | TruG-TRBM · Bouncing Ball | test prediction error: 4.16±0.42 -> 3.66±0.46 | Improved |
| Use fixed ReLU-type truncation points rather than tanh-type truncation points (ablation) | TruG-TRBM · CMU Motion Capture | test prediction error: 21.8±0.31 vs 7.3±0.21; 14.9±0.29 vs 5.9±0.22 | Mixed |
| Learn truncation points instead of fixing them to a standard nonlinearity (ablation) | TruG-TGGM · UCI regression datasets | RMSE | Mixed |
| Use unit-specific rather than shared learned truncation points (ablation) | TruG-TGGM · UCI regression datasets | RMSE | Mixed |
Closest papers by meaning
- Nonlinear Statistical Learning with Truncated Gaussian Graphical Models (ICML 2016 · similarity 0.87)
- Inferring stochastic low-rank recurrent neural networks from neural data (NeurIPS 2024 · similarity 0.85)
- Asymptotics of feature learning in two-layer networks after one gradient-step (ICML 2024 · similarity 0.84)
- Stochastic Modified Equations and Dynamics of Dropout Algorithm (ICLR 2024 · similarity 0.84)
- The Power of Random Features and the Limits of Distribution-Free Gradient Descent (ICML 2025 · similarity 0.84)
- Learning with Restricted Boltzmann Machines: Asymptotics of AMP and GD in High Dimensions (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.
