Papers
Topics
Authors
Recent
Search
2000 character limit reached

Correlation of Loss Differences (CLD)

Updated 9 July 2026
  • CLD is defined as a coreset selection metric that quantifies how a training sample’s loss trajectory correlates with its class-average validation loss trajectory.
  • It efficiently selects representative samples by relying solely on forward-pass loss differences, avoiding expensive gradient or curvature computations.
  • Empirical results on datasets like CIFAR-100 and ImageNet-1k demonstrate that CLD coresets achieve near state-of-the-art accuracy with reduced computational overhead and strong transferability.

66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66^ 66 66 OR \666 66 66 66 66 66 66 66 66 66 66^ 66^ 66 66 OR \666 66 66^ 66^ 66 66 66 66 66 66 66^ 66^ 66 66 OR \666 66 66^ 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66^ 66 66 OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR \666 OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 coreset666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 OR \666 66 66^ 66 66^ 66 66 66 66 66 from Trajectories: Selecting Data via Correlation of Loss Differences,66^ 66 66 66 Noise Induces Loss Discrepancy Across Groups66, 66 66 66 66 66^ 66 the correlation between operational risk loss categories over different time horizons66^ 66^ 66 66 OR \666 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 coreset666 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 coreset666 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 from Trajectories: Selecting Data via Correlation of Loss Differences,66^ 66^ 66 66 Noise Induces Loss Discrepancy Across Groups66^ 66 66 66 66 66 the correlation between operational risk loss categories over different time horizons66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR \666 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 coreset666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 OR \666 OR \666 66 66 OR \666 OR \666 66 66 OR \666 OR \666 OR \666 OR \666 OR \666 66 66 66 OR \666 OR \666 OR \666 OR \666 66 66 66 66 66 66 66 OR \666 OR \666 OR \666 66 66 66 Noise Induces Loss Discrepancy Across Groups66M 66 66 66 OR \666 OR \666 OR \666 OR \666 OR \666 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR \666 66 66 66 66 66 66 66 66 66) OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR \666 OR \666 66 66 66 66 66 66 66 66 66 66Estimating the correlation between operational risk loss categories over different time horizons66 OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 OR \666 OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 OR \666 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 Noise Induces Loss Discrepancy Across Groups66^ 66 66 66 66 66 66 66 66 OR \666 OR \666 66 66 66 the correlation between operational risk loss categories over different time horizons66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66 66 66 66 66 66 66Coresets from Trajectories: Selecting Data via Correlation of Loss Differences,66 OR \666 66 66 66 OR \666 OR \666 66 66 66 66 66 66 66 Noise Induces Loss Discrepancy Across Groups66^ 66 66Coresets from Trajectories: Selecting Data via Correlation of Loss Differences,66 coreset666 OR \666 66 66Coresets from Trajectories: Selecting Data via Correlation of Loss Differences,66 coreset666 OR \666 from Trajectories: Selecting Data via Correlation of Loss Differences,66% 66 66 66 66) OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 coreset666 OR \666 66 66Estimating the correlation between operational risk loss categories over different time horizons66 OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 from Trajectories: Selecting Data via Correlation of Loss Differences,66^ 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66^ 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66) OR \666 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66 66

Topic to Video (Beta)

No one has generated a video about this topic yet.

Sign Up to Generate All Videos Subscribe on YouTube

Whiteboard

No one has generated a whiteboard explanation for this topic yet.

Sign Up to Generate

Follow Topic

Get notified by email when new papers are published related to Correlation of Loss Differences (CLD).

Sign Up to Follow Topic by Email