Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Benchmarking Bayesian Deep Learning on Diabetic Retinopathy Detection Tasks
Neil Band, Tim G. J. Rudner, Qixuan Feng, Angelos Filos, Zachary Nado, Michael W Dusenberry, Ghassen Jerfel, Dustin Tran, Yarin Gal
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 |
|---|---|---|---|
| Use MC Dropout uncertainty for selective referral instead of MAP uncertainty | Retina benchmark models · Severity Shift: severe/proliferative | referral rate at 100% accuracy: nearly 20% lower | Improved |
| Use uncertainty-based referral as referral rate increases | most methods · Country Shift: APTOS | AUC | Hurt |
| Use uncertainty-based referral with MFVI | MFVI · Country Shift: APTOS | AUC: 91.4% up to 93.8% | Mixed |
| Add epistemic uncertainty using MC Dropout rather than deterministic MAP | Retina models · Severity Shift: severe/proliferative | referral rate needed for 100% accuracy: nearly 20% lower | Improved |
| Use variational inference methods, including MC Dropout, FSVI, and Deep Ensemble, instead of MAP | Retina benchmark models · Country Shift: in-domain | AUC | Improved |
| Tune hyperparameters on balanced joint in-domain and shifted validation data instead of in-domain validation AUC (ablation) | Retina benchmark models · Country Shift: APTOS | accuracy referral curve | Improved |
| Use a single Monte Carlo sample at training time, enabling larger batch sizes (ablation) | variational methods · Retina benchmark validation | validation performance | Improved |
| Use constant class weights instead of mini-batch-relative class weighting (ablation) | Retina benchmark training · diabetic retinopathy detection | overall performance | Hurt |
Closest papers by meaning
- FOCUS: Benchmarking Retinal Model Generalization from Foundation Vision Encoders to Multimodal LLMs (NeurIPS 2026 · similarity 0.91)
- Learning model uncertainty as variance-minimizing instance weights (ICLR 2024 · similarity 0.87)
- HACDR-Net: Heterogeneous-Aware Convolutional Network for Diabetic Retinopathy Multi-Lesion Segmentation (AAAI 2024 · similarity 0.85)
- Like an Ophthalmologist: Dynamic Selection Driven Multi-View Learning for Diabetic Retinopathy Grading (AAAI 2025 · similarity 0.85)
- Domain Adaptive Diabetic Retinopathy Grading with Model Absence and Flowing Data (CVPR 2025 · similarity 0.84)
- GlucoFM-Bench: Benchmarking Time-Series Foundation Models for Blood Glucose Forecasting (NeurIPS 2026 · 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.
