AI RESEARCH
TopoPrune: Robust Data Pruning via Unified Latent Space Topology
arXiv CS.AI
•
ArXi:2602.02739v2 Announce Type: replace-cross Geometric data pruning methods, while practical for leveraging pretrained models, are fundamentally unstable. Their reliance on extrinsic geometry renders them highly sensitive to latent space perturbations, causing performance to degrade during cross-architecture transfer or in the presence of feature noise. We