Researchexpectation maximizationcomputer visionzalandohuman in the loop
Zalando Introduces AutoThreshold For Classifier Calibration
|
8.2
Zalando describes AutoThreshold, an EM-based algorithm that tunes class-confidence thresholds for image classifiers using iterative human annotations on out-of-distribution product images. The method samples high-confidence predictions for manual labeling, optimizes f_beta-based thresholds across iterations, and builds a compact validation set to mitigate data drift and improve search relevance at scale (Zalando cites about 600,000 products).
Scoring Rationale
Practical EM-based thresholding offers actionable drift mitigation, with strength in applied detail but limited external validation beyond a company blog.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
