Research map / Image generation / Fast diffusion sampling
Robustness of Probabilistic Models to Low-Quality Data: A Multi-Perspective Analysis
Liu Peng, Yaochu Jin
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 100% incorrect labels (error rate 50%) (ablation) | EDM class-conditional diffusion · CIFAR-10 generation | label consistency accuracy: 94.082% -> 40.630% | Hurt |
| add 100% incorrect labels (error rate 50%) (ablation) | EDM class-conditional diffusion · CIFAR-100 generation | label consistency accuracy: 65.236% -> 16.996% | Hurt |
| add 50% corrupted tokens (ablation) | GPT-2 124M · OpenWebText | test NLL: 2.87 -> 3.59 | Hurt |
| add 100% incorrect labels (ablation) | ResNet-18 · CIFAR-10 classification | accuracy: 95.30% -> 85.35% | Hurt |
| add 100% incorrect labels (ablation) | ViT-Base · ImageNet-100 | accuracy: 64.520% -> 45.920% | Hurt |
| add 100% incorrect labels (ablation) | ViT-Base · ImageNet-1000 | accuracy: 73.784% -> 74.778% | No clear effect |
| richer conditioning input (CNN/DailyMail vs WMT) | seq2seq models trained with noisy teacher targets · 50% error rate | relative NLL increase: 31.5% -> 17.9% | Improved |
| increase batch size under high noise (ablation) | gradient aggregation · gradient analysis, 50% corruption | signal-to-noise ratio: 2.96x -> 3.83x | Improved |
| add incorrect labels (ablation) | EDM class-conditional diffusion · CIFAR generation | FID | No clear effect |
Closest papers by meaning
- Remarkable Robustness of LLMs: Stages of Inference? (NeurIPS 2025 · similarity 0.88)
- Brittlebench: Quantifying LLM robustness via prompt sensitivity (NeurIPS 2026 · similarity 0.87)
- An Empirical Study on Noisy Data and LLM Pretraining Loss Divergence (NeurIPS 2026 · similarity 0.87)
- A Probabilistic Perspective on Unlearning and Alignment for Large Language Models (ICLR 2025 · similarity 0.87)
- Signal and Noise: A Framework for Reducing Uncertainty in Language Model Evaluation (NeurIPS 2025 · similarity 0.87)
- Understanding Domain Generalization: A Noise Robustness Perspective (ICLR 2024 · similarity 0.87)
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.
