Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Towards Knowledge-augmented Bayesian Deep Learning For Computer Vision
Wang Ma, Hanjing Wang, Yufei Zhang, Darsha Udayanga, Qiang Ji
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 Stage 1 knowledge-informed prior on top of adaptive likelihood (ablation) | knowledge-augmented BDL · Decoy MNIST | NLL: 0.054 -> 0.043 | Improved |
| add Stage 1 knowledge-informed prior on top of adaptive likelihood (ablation) | knowledge-augmented BDL · Rotated MNIST | accuracy (%): 94.45 -> 94.68 | Improved |
| add Stage 1 knowledge-informed prior on top of adaptive likelihood (ablation) | knowledge-augmented BDL · FreiHAND 3D hand reconstruction | knowledge constraint KC: 0.26 -> 0.02 | Improved |
| add Stage 1 knowledge-informed prior on top of adaptive likelihood (ablation) | knowledge-augmented BDL · FreiHAND 3D hand reconstruction | joint error EJ (mm): 8.01 -> 7.89 | Improved |
| add Stage 1 knowledge-informed prior (ablation) | knowledge-augmented BDL · FreiHAND angle violation rate | angle violation rate (%): 2.57% -> 0.14% | Improved |
| add adaptive knowledge likelihood during Stage 2 instead of prior-only knowledge (BANANA) | Bayesian deep ensemble · Rotated MNIST | accuracy (%): 51.11 -> 94.45 | Improved |
| add adaptive knowledge likelihood vs BANANA prior-only | Bayesian deep ensemble · FreiHAND 3D hand reconstruction | joint error EJ (mm): 19.83 -> 8.01 | Improved |
| add Stage 1 prior (ablation) | knowledge-augmented BDL · Decoy MNIST with 1% to 10% training data | accuracy | Improved |
| add Stage 1 prior (ablation) | knowledge-augmented BDL · Decoy MNIST to KMNIST OOD | AUROC (%): 96.78 -> 97.10 | Mixed |
| change non-informative prior of likelihood-only model from Gaussian to Laplacian or Logistic (ablation) | Ours (Likelihood-Only) · Decoy MNIST | accuracy (%): 98.33 -> 98.30 (Laplacian), 98.39 (Logistic) | No clear effect |
| swap ensemble posterior estimator for SGLD, Bayes-Backprop or Laplace approximation (ablation) | Ours (Likelihood-Only) · Decoy MNIST | accuracy (%): 98.33 -> 98.45 (SGLD), 97.76 (Bayes-Backprop), 97.82 (LA) | No clear effect |
| more iterations of adaptive likelihood (Alg. 1) (ablation) | Ours (Likelihood-Only) · Decoy MNIST | accuracy (%): 97.52 -> 98.28 | Improved |
Closest papers by meaning
- Position: Bayesian Deep Learning is Needed in the Age of Large-Scale AI (ICML 2024 · similarity 0.90)
- Practical Deep Learning with Bayesian Principles (NeurIPS 2019 · similarity 0.87)
- Bayesian Exploration of Pre-trained Models for Low-shot Image Classification (CVPR 2024 · similarity 0.86)
- Make Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models (CVPR 2024 · similarity 0.85)
- Bayesian Test-Time Adaptation for Vision-Language Models (CVPR 2025 · similarity 0.85)
- Bayesian Low-Rank Learning (Bella): A Practical Approach to Bayesian Neural Networks (AAAI 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.
