Dataset Pruning from First Principles: A Label-Free Linear Programming Approach

Key points

• Existing geometry-based methods typically assume that nearby points in embedding space share similar properties.1
• Rather than imposing this assumption, we derive geometric selection criteria by reformulating unbiased subset selection as a variance minimization problem.1
• Specifically, we characterize a family of unbiased subset selection algorithms as a high-dimensional polytope.1
• We instead use these expressions to construct an efficient vertex walk that optimizes an approximation of the variance objective while preserving unbiasedness, yielding a method that requires neither labels nor model training during selection.1

Sources (1)

• [1]Dataset Pruning from First Principles: A Label-Free Linear Programming Approach
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 04:24 PM

  Dataset pruning reduces a large training set to a representative subset while preserving model performance.

  Existing geometry-based methods typically assume that nearby points in embedding space share similar properties.

Extractive summary: sentences quoted from the sources.