Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Dichotomize and Generalize: PAC-Bayesian Binary Activated Deep Neural Networks
Gaël Letarte, Pascal Germain, Benjamin Guedj, Francois Laviolette
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 pretraining on half the training data and use the learned weights as the prior (ablation) | PBGNet · MNIST 4 vs 9 | test error: 0.040 -> 0.017 | Improved |
| Add pretraining on half the training data and use the learned weights as the prior (ablation) | PBGNet · MNIST 5 vs 6 | test error: 0.025 -> 0.009 | Improved |
| Add pretraining on half the training data and use the learned weights as the prior (ablation) | PBGNet · Ads | test error: 0.168 -> 0.033 | Improved |
| Add pretraining on half the training data and use the learned weights as the prior (ablation) | PBGNet · Adult | test error: 0.163 -> 0.154 | Improved |
| Add pretraining on half the training data and use the learned weights as the prior (ablation) | PBGNet · MNIST low vs high | test error: 0.052 -> 0.027 | Improved |
| Select hyperparameters using the bound while training on empirical loss (ablation) | PBGNet_ℓ · MNIST 4 vs 9 | test error: 0.018 -> 0.035 | Hurt |
| Optimize the PAC-Bayes bound directly rather than empirical loss (ablation) | PBGNet · Ads | test error: 0.030 -> 0.168 | Hurt |
| Optimize the PAC-Bayes bound directly rather than empirical loss (ablation) | PBGNet · Adult | test error: 0.149 -> 0.163 | Hurt |
| Optimize the PAC-Bayes bound directly rather than empirical loss (ablation) | PBGNet · MNIST 4 vs 9 | test error: 0.018 -> 0.040 | Hurt |
| Reduce Monte Carlo sample size during training (ablation) | PBGNet · some learning problems | training performance | Mixed |
Closest papers by meaning
- Deterministic PAC-Bayesian generalization bounds for deep networks via generalizing noise-resilience (ICLR 2019 · similarity 0.90)
- A PAC-Bayesian Approach to Generalization Bounds for Graph Neural Networks (ICLR 2021 · similarity 0.87)
- Certifying Deep Network Risks and Individual Predictions with PAC-Bayes Loss via Localized Priors (NeurIPS 2025 · similarity 0.87)
- Deep learning generalizes because the parameter-function map is biased towards simple functions (ICLR 2019 · similarity 0.86)
- Non-Vacuous Generalisation Bounds for Shallow Neural Networks (ICML 2022 · similarity 0.86)
- Generalization Bounds via Meta-Learned Model Representations: PAC-Bayes and Sample Compression Hypernetworks (ICML 2025 · similarity 0.85)
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.
