arXiv:2605.10165v2 Announce Type: replace-cross Abstract: Noisy labels are common in large-scale medical imaging datasets due to inter-observer variability and ambiguous cases. We propose a statistically grounded and task-agnostic framework, Standardized Loss Aggregation (SLA), for detecting noisy labels at the sample level. SLA quantifies label reliability by aggregating standardized fold-level validation losses across repeated cross-validation runs. This formulation generalizes discrete hard-counting schemes into a continuous estimator that captures both the frequency and magnitude of perfor
The proliferation of large-scale AI datasets, particularly in sensitive domains like medical imaging, exacerbates the challenge of noisy labels, making robust detection methods critical.
Improved detection of noisy labels in AI training datasets enhances model reliability and trustworthiness, crucial for deploying AI in high-stakes environments.
This framework offers a statistically grounded approach to automate the identification of unreliable data points, reducing manual effort and improving dataset quality.
- · AI developers
- · Healthcare sector
- · Data scientists
- · AI ethics and safety
- · Companies with poor data labeling practices
AI models trained on cleaner data will exhibit higher accuracy and robustness.
The cost and time associated with manual data curation could decrease, accelerating AI development cycles.
Increased trust in AI systems due to improved data quality could lead to wider adoption in critical applications.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.AI