Zobrazeno 1 - 10
of 19
pro vyhledávání: '"Wutschitz, Lukas"'
Autor:
Siddiqui, Shoaib Ahmed, Gaonkar, Radhika, Köpf, Boris, Krueger, David, Paverd, Andrew, Salem, Ahmed, Tople, Shruti, Wutschitz, Lukas, Xia, Menglin, Zanella-Béguelin, Santiago
Large Language Models (LLMs) are rapidly becoming commodity components of larger software systems. This poses natural security and privacy problems: poisoned data retrieved from one component can change the model's behavior and compromise the entire
Externí odkaz:
http://arxiv.org/abs/2410.03055
Autor:
Cherubin, Giovanni, Köpf, Boris, Paverd, Andrew, Tople, Shruti, Wutschitz, Lukas, Zanella-Béguelin, Santiago
Machine learning models trained with differentially-private (DP) algorithms such as DP-SGD enjoy resilience against a wide range of privacy attacks. Although it is possible to derive bounds for some attacks based solely on an $(\varepsilon,\delta)$-D
Externí odkaz:
http://arxiv.org/abs/2402.14397
Autor:
Lukas, Nils, Salem, Ahmed, Sim, Robert, Tople, Shruti, Wutschitz, Lukas, Zanella-Béguelin, Santiago
Language Models (LMs) have been shown to leak information about training data through sentence-level membership inference and reconstruction attacks. Understanding the risk of LMs leaking Personally Identifiable Information (PII) has received less at
Externí odkaz:
http://arxiv.org/abs/2302.00539
Autor:
Zanella-Béguelin, Santiago, Wutschitz, Lukas, Tople, Shruti, Salem, Ahmed, Rühle, Victor, Paverd, Andrew, Naseri, Mohammad, Köpf, Boris, Jones, Daniel
Algorithms such as Differentially Private SGD enable training machine learning models with formal privacy guarantees. However, there is a discrepancy between the protection that such algorithms guarantee in theory and the protection they afford in pr
Externí odkaz:
http://arxiv.org/abs/2206.05199
Autor:
Mireshghallah, Fatemehsadat, Backurs, Arturs, Inan, Huseyin A, Wutschitz, Lukas, Kulkarni, Janardhan
Recent papers have shown that large pre-trained language models (LLMs) such as BERT, GPT-2 can be fine-tuned on private data to achieve performance comparable to non-private models for many downstream Natural Language Processing (NLP) tasks while sim
Externí odkaz:
http://arxiv.org/abs/2206.01838
Autor:
Yu, Da, Naik, Saurabh, Backurs, Arturs, Gopi, Sivakanth, Inan, Huseyin A., Kamath, Gautam, Kulkarni, Janardhan, Lee, Yin Tat, Manoel, Andre, Wutschitz, Lukas, Yekhanin, Sergey, Zhang, Huishuai
We give simpler, sparser, and faster algorithms for differentially private fine-tuning of large-scale pre-trained language models, which achieve the state-of-the-art privacy versus utility tradeoffs on many standard NLP tasks. We propose a meta-frame
Externí odkaz:
http://arxiv.org/abs/2110.06500
We give a fast algorithm to optimally compose privacy guarantees of differentially private (DP) algorithms to arbitrary accuracy. Our method is based on the notion of privacy loss random variables to quantify the privacy loss of DP algorithms. The ru
Externí odkaz:
http://arxiv.org/abs/2106.02848
Autor:
Inan, Huseyin A., Ramadan, Osman, Wutschitz, Lukas, Jones, Daniel, Rühle, Victor, Withers, James, Sim, Robert
Recent advances in neural network based language models lead to successful deployments of such models, improving user experience in various applications. It has been demonstrated that strong performance of language models comes along with the ability
Externí odkaz:
http://arxiv.org/abs/2101.05405
Autor:
Zanella-Béguelin, Santiago, Wutschitz, Lukas, Tople, Shruti, Rühle, Victor, Paverd, Andrew, Ohrimenko, Olga, Köpf, Boris, Brockschmidt, Marc
To continuously improve quality and reflect changes in data, machine learning applications have to regularly retrain and update their core models. We show that a differential analysis of language model snapshots before and after an update can reveal
Externí odkaz:
http://arxiv.org/abs/1912.07942
Publikováno v:
In Journal of Computational Physics 1 May 2020 408
