Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Epistemic Uncertainty Quantification for Pre-trained VLMs via Riemannian Flow Matching
Li Ju, Mayank Nautiyal, Andreas Hellander, Ekta Vats, Prashant Singh
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 Riemannian (geodesic) flow matching instead of Euclidean variants (ablation) | REPVLM · ImageNet-1K, Conceptual Captions proxy | accuracy-rejection curve | Improved |
| increase proxy dataset size from 50K to 1M pairs (ablation) | REPVLM · ImageNet-1K, Conceptual Captions proxy | Acc at 90% rejection | Improved |
| increase proxy dataset beyond 500K samples (ablation) | REPVLM · ImageNet-1K, Conceptual Captions proxy | Acc at 90% rejection | No clear effect |
| increase ODE integration steps from 1-2 to 5 (ablation) | REPVLM · ImageNet-1K, Conceptual Captions proxy | accuracy-rejection curve | Improved |
| increase ODE steps from 5 to 8 (ablation) | REPVLM · ImageNet-1K, Conceptual Captions proxy | accuracy-rejection curve | No clear effect |
| use embedding density instead of MC dropout / ProbVLM for epistemic UQ | REPVLM · ImageNet-1K, Conceptual Captions proxy | Acc@90% rejection: 0.710 -> 0.733 (vs MC-Dropout) | Improved |
| use REPVLM instead of embedding norm on Food101 | REPVLM · Food101 | Acc@90% rejection: 0.934 (Emb. Norm) vs 0.904 | Hurt |
| use REPVLM instead of Conformity on ImageNet-Sketch | REPVLM · ImageNet-Sketch | Acc@90% rejection: 0.672 (Conformity) vs 0.656 (CCaps proxy) | Hurt |
| train on noisy LAION proxy instead of cleaner proxies | REPVLM · ImageNet-Sketch | Acc@90% rejection: 0.656 -> 0.618 | Hurt |
| use density-based scoring with REPVLM versus a Euclidean GMM density | REPVLM vs GMM · Food101 | Spearman S: -0.479 -> 1.000 | Improved |
| use REPVLM with 5 ODE steps instead of MC dropout with 10 passes | REPVLM · Computational cost | GFLOPs: 92.50 -> 9.80 | Improved |
Closest papers by meaning
- Exploiting the Asymmetric Uncertainty Structure of Pre-trained VLMs on the Unit Hypersphere (NeurIPS 2025 · similarity 0.90)
- FUSE: Quantifying Uncertainty in Vision-Language Models by Bayesian Fusing Epistemic and Aleatoric Uncertainty (ICML 2026 · similarity 0.89)
- CertainlyUncertain: A Benchmark and Metric for Multimodal Epistemic and Aleatoric Awareness (ICLR 2025 · similarity 0.88)
- QUANTIPHY: A Quantitative Benchmark Evaluating Physical Reasoning Abilities of Vision-Language Models (CVPR 2026 · similarity 0.86)
- Enhancing Vision-Language Model Reliability with Uncertainty-Guided Dropout Decoding (NeurIPS 2025 · similarity 0.86)
- Probabilistic Language-Image Pre-Training (ICLR 2025 · similarity 0.86)
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.
