y0news
← Feed
Back to feed
🧠 AI🟢 Bullish

Towards Principled Dataset Distillation: A Spectral Distribution Perspective

arXiv – CS AI|Ruixi Wu, Shaobo Wang, Jiahuan Chen, Zhiyuan Liu, Yicun Yang, Zhaorun Chen, Zekai Li, Kaixin Li, Xinming Wang, Hongzhu Yi, Kai Wang, Linfeng Zhang||1 views
🤖AI Summary

Researchers propose Class-Aware Spectral Distribution Matching (CSDM), a new dataset distillation method that addresses performance issues on imbalanced datasets. The technique achieves 14% improvement over existing methods on CIFAR-10-LT with enhanced stability on long-tailed data distributions.

Key Takeaways
  • CSDM reformulates distribution alignment using spectral analysis of kernel functions to improve dataset distillation performance.
  • The method specifically addresses challenges with long-tailed datasets where existing approaches show substantial performance degradation.
  • Amplitude-phase decomposition is used to adaptively prioritize realism in underrepresented tail classes.
  • Testing on CIFAR-10-LT shows 14% improvement over state-of-the-art methods with only 5.7% performance drop on extreme imbalance.
  • The approach demonstrates strong stability when tail class samples decrease dramatically from 500 to 25 images.
Read Original →via arXiv – CS AI
Act on this with AI
Stay ahead of the market.
Connect your wallet to an AI agent. It reads balances, proposes swaps and bridges across 15 chains — you keep full control of your keys.
Connect Wallet to AI →How it works
Related Articles