Search papers, labs, and topics across Lattice.
3
0
7
Pruning LLMs while preserving their influence functions is now over 28x faster and more effective, unlocking better data selection for downstream tasks.
Non-private synthetic data can effectively transfer knowledge from original corpora, while state-of-the-art DP methods often fail to do so, even at high privacy levels.
By strategically applying differential privacy in the wavelet domain, this work achieves a significantly better privacy-utility tradeoff for image generation, suggesting that not all image frequencies are created equal when it comes to privacy risk.
