Zobrazeno 1 - 10
of 505
pro vyhledávání: '"Zhang, Chiyuan"'
Autor:
Shi, Weijia, Lee, Jaechan, Huang, Yangsibo, Malladi, Sadhika, Zhao, Jieyu, Holtzman, Ari, Liu, Daogao, Zettlemoyer, Luke, Smith, Noah A., Zhang, Chiyuan
Language models (LMs) are trained on vast amounts of text data, which may include private and copyrighted content. Data owners may request the removal of their data from a trained model due to privacy or copyright concerns. However, exactly unlearnin
Externí odkaz:
http://arxiv.org/abs/2407.06460
We study the differentially private (DP) empirical risk minimization (ERM) problem under the semi-sensitive DP setting where only some features are sensitive. This generalizes the Label DP setting where only the label is sensitive. We give improved u
Externí odkaz:
http://arxiv.org/abs/2406.19040
Autor:
Wei, Boyi, Shi, Weijia, Huang, Yangsibo, Smith, Noah A., Zhang, Chiyuan, Zettlemoyer, Luke, Li, Kai, Henderson, Peter
Language models (LMs) derive their capabilities from extensive training on diverse data, including potentially copyrighted material. These models can memorize and generate content similar to their training data, posing potential concerns. Therefore,
Externí odkaz:
http://arxiv.org/abs/2406.18664
Autor:
Chua, Lynn, Ghazi, Badih, Huang, Yangsibo, Kamath, Pritish, Kumar, Ravi, Manurangsi, Pasin, Sinha, Amer, Xie, Chulin, Zhang, Chiyuan
Large language models (LLMs) are typically multilingual due to pretraining on diverse multilingual corpora. But can these models relate corresponding concepts across languages, effectively being crosslingual? This study evaluates six state-of-the-art
Externí odkaz:
http://arxiv.org/abs/2406.16135
Autor:
He, Luxi, Huang, Yangsibo, Shi, Weijia, Xie, Tinghao, Liu, Haotian, Wang, Yue, Zettlemoyer, Luke, Zhang, Chiyuan, Chen, Danqi, Henderson, Peter
Recent studies show that image and video generation models can be prompted to reproduce copyrighted content from their training data, raising serious legal concerns around copyright infringement. Copyrighted characters, in particular, pose a difficul
Externí odkaz:
http://arxiv.org/abs/2406.14526
Autor:
Chua, Lynn, Ghazi, Badih, Huang, Yangsibo, Kamath, Pritish, Kumar, Ravi, Liu, Daogao, Manurangsi, Pasin, Sinha, Amer, Zhang, Chiyuan
Large language models (LLMs) have emerged as powerful tools for tackling complex tasks across diverse domains, but they also raise privacy concerns when fine-tuned on sensitive data due to potential memorization. While differential privacy (DP) offer
Externí odkaz:
http://arxiv.org/abs/2406.14322
Can we localize the weights and mechanisms used by a language model to memorize and recite entire paragraphs of its training data? In this paper, we show that while memorization is spread across multiple layers and model components, gradients of memo
Externí odkaz:
http://arxiv.org/abs/2403.19851
Autor:
Chua, Lynn, Ghazi, Badih, Kamath, Pritish, Kumar, Ravi, Manurangsi, Pasin, Sinha, Amer, Zhang, Chiyuan
We demonstrate a substantial gap between the privacy guarantees of the Adaptive Batch Linear Queries (ABLQ) mechanism under different types of batch sampling: (i) Shuffling, and (ii) Poisson subsampling; the typical analysis of Differentially Private
Externí odkaz:
http://arxiv.org/abs/2403.17673
Autor:
Chua, Lynn, Cui, Qiliang, Ghazi, Badih, Harrison, Charlie, Kamath, Pritish, Krichene, Walid, Kumar, Ravi, Manurangsi, Pasin, Narra, Krishna Giri, Sinha, Amer, Varadarajan, Avinash, Zhang, Chiyuan
Motivated by problems arising in digital advertising, we introduce the task of training differentially private (DP) machine learning models with semi-sensitive features. In this setting, a subset of the features is known to the attacker (and thus nee
Externí odkaz:
http://arxiv.org/abs/2401.15246
Autor:
Badanidiyuru, Ashwinkumar, Ghazi, Badih, Kamath, Pritish, Kumar, Ravi, Leeman, Ethan, Manurangsi, Pasin, Varadarajan, Avinash V, Zhang, Chiyuan
We propose a new family of label randomizers for training regression models under the constraint of label differential privacy (DP). In particular, we leverage the trade-offs between bias and variance to construct better label randomizers depending o
Externí odkaz:
http://arxiv.org/abs/2312.05659