@article{ZieglerPfitznerSchulzetal.2022,
  author    = {Ziegler, Joceline and Pfitzner, Bjarne and Schulz, Heinrich and Saalbach, Axel and Arnrich, Bert},
  title     = {Defending against Reconstruction Attacks through Differentially Private Federated Learning for Classification of Heterogeneous Chest X-ray Data},
  series = {Sensors},
  volume    = {22},
  journal   = {Sensors},
  edition   = {14},
  publisher = {MDPI},
  address   = {Basel, Schweiz},
  issn      = {1424-8220},
  doi       = {10.3390/s22145195},
  pages     = {25},
  year      = {2022},
  abstract  = {Privacy regulations and the physical distribution of heterogeneous data are often primary concerns for the development of deep learning models in a medical context. This paper evaluates the feasibility of differentially private federated learning for chest X-ray classification as a defense against data privacy attacks. To the best of our knowledge, we are the first to directly compare the impact of differentially private training on two different neural network architectures, DenseNet121 and ResNet50. Extending the federated learning environments previously analyzed in terms of privacy, we simulated a heterogeneous and imbalanced federated setting by distributing images from the public CheXpert and Mendeley chest X-ray datasets unevenly among 36 clients. Both non-private baseline models achieved an area under the receiver operating characteristic curve (AUC) of 0.940.94 on the binary classification task of detecting the presence of a medical finding. We demonstrate that both model architectures are vulnerable to privacy violation by applying image reconstruction attacks to local model updates from individual clients. The attack was particularly successful during later training stages. To mitigate the risk of a privacy breach, we integrated R{\´e}nyi differential privacy with a Gaussian noise mechanism into local model training. We evaluate model performance and attack vulnerability for privacy budgets ε∈{1,3,6,10}�∈{1,3,6,10}. The DenseNet121 achieved the best utility-privacy trade-off with an AUC of 0.940.94 for ε=6�=6. Model performance deteriorated slightly for individual clients compared to the non-private baseline. The ResNet50 only reached an AUC of 0.760.76 in the same privacy setting. Its performance was inferior to that of the DenseNet121 for all considered privacy constraints, suggesting that the DenseNet121 architecture is more robust to differentially private training.},
  language  = {en}
}
@misc{ZieglerPfitznerSchulzetal.2022,
  author    = {Ziegler, Joceline and Pfitzner, Bjarne and Schulz, Heinrich and Saalbach, Axel and Arnrich, Bert},
  title     = {Defending against Reconstruction Attacks through Differentially Private Federated Learning for Classification of Heterogeneous Chest X-ray Data},
  series = {Zweitver{\"o}ffentlichungen der Universit{\"a}t Potsdam : Reihe der Digital Engineering Fakult{\"a}t},
  journal   = {Zweitver{\"o}ffentlichungen der Universit{\"a}t Potsdam : Reihe der Digital Engineering Fakult{\"a}t},
  number    = {14},
  doi       = {10.25932/publishup-58132},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-581322},
  pages     = {25},
  year      = {2022},
  abstract  = {Privacy regulations and the physical distribution of heterogeneous data are often primary concerns for the development of deep learning models in a medical context. This paper evaluates the feasibility of differentially private federated learning for chest X-ray classification as a defense against data privacy attacks. To the best of our knowledge, we are the first to directly compare the impact of differentially private training on two different neural network architectures, DenseNet121 and ResNet50. Extending the federated learning environments previously analyzed in terms of privacy, we simulated a heterogeneous and imbalanced federated setting by distributing images from the public CheXpert and Mendeley chest X-ray datasets unevenly among 36 clients. Both non-private baseline models achieved an area under the receiver operating characteristic curve (AUC) of 0.940.94 on the binary classification task of detecting the presence of a medical finding. We demonstrate that both model architectures are vulnerable to privacy violation by applying image reconstruction attacks to local model updates from individual clients. The attack was particularly successful during later training stages. To mitigate the risk of a privacy breach, we integrated R{\´e}nyi differential privacy with a Gaussian noise mechanism into local model training. We evaluate model performance and attack vulnerability for privacy budgets ε∈{1,3,6,10}�∈{1,3,6,10}. The DenseNet121 achieved the best utility-privacy trade-off with an AUC of 0.940.94 for ε=6�=6. Model performance deteriorated slightly for individual clients compared to the non-private baseline. The ResNet50 only reached an AUC of 0.760.76 in the same privacy setting. Its performance was inferior to that of the DenseNet121 for all considered privacy constraints, suggesting that the DenseNet121 architecture is more robust to differentially private training.},
  language  = {en}
}
@article{PfitznerSteckhanArnrich2021,
  author    = {Pfitzner, Bjarne and Steckhan, Nico and Arnrich, Bert},
  title     = {Federated learning in a medical context},
  series = {ACM transactions on internet technology : TOIT / Association for Computing},
  volume    = {21},
  journal   = {ACM transactions on internet technology : TOIT / Association for Computing},
  number    = {2},
  publisher = {Association for Computing Machinery},
  address   = {New York},
  issn      = {1533-5399},
  doi       = {10.1145/3412357},
  pages     = {1 -- 31},
  year      = {2021},
  abstract  = {Data privacy is a very important issue. Especially in fields like medicine, it is paramount to abide by the existing privacy regulations to preserve patients' anonymity. However, data is required for research and training machine learning models that could help gain insight into complex correlations or personalised treatments that may otherwise stay undiscovered. Those models generally scale with the amount of data available, but the current situation often prohibits building large databases across sites. So it would be beneficial to be able to combine similar or related data from different sites all over the world while still preserving data privacy. Federated learning has been proposed as a solution for this, because it relies on the sharing of machine learning models, instead of the raw data itself. That means private data never leaves the site or device it was collected on. Federated learning is an emerging research area, and many domains have been identified for the application of those methods. This systematic literature review provides an extensive look at the concept of and research into federated learning and its applicability for confidential healthcare datasets.},
  language  = {en}
}
@book{KubanRottaNolteetal.2023,
  author    = {Kuban, Robert and Rotta, Randolf and Nolte, J{\"o}rg and Chromik, Jonas and Beilharz, Jossekin Jakob and Pirl, Lukas and Friedrich, Tobias and Lenzner, Pascal and Weyand, Christopher and Juiz, Carlos and Bermejo, Belen and Sauer, Joao and Coelh, Leandro dos Santos and Najafi, Pejman and P{\"u}nter, Wenzel and Cheng, Feng and Meinel, Christoph and Sidorova, Julia and Lundberg, Lars and Vogel, Thomas and Tran, Chinh and Moser, Irene and Grunske, Lars and Elsaid, Mohamed Esameldin Mohamed and Abbas, Hazem M. and Rula, Anisa and Sejdiu, Gezim and Maurino, Andrea and Schmidt, Christopher and H{\"u}gle, Johannes and Uflacker, Matthias and Nozza, Debora and Messina, Enza and Hoorn, Andr{\´e} van and Frank, Markus and Schulz, Henning and Alhosseini Almodarresi Yasin, Seyed Ali and Nowicki, Marek and Muite, Benson K. and Boysan, Mehmet Can and Bianchi, Federico and Cremaschi, Marco and Moussa, Rim and Abdel-Karim, Benjamin M. and Pfeuffer, Nicolas and Hinz, Oliver and Plauth, Max and Polze, Andreas and Huo, Da and Melo, Gerard de and Mendes Soares, F{\´a}bio and Oliveira, Roberto C{\´e}lio Lim{\~a}o de and Benson, Lawrence and Paul, Fabian and Werling, Christian and Windheuser, Fabian and Stojanovic, Dragan and Djordjevic, Igor and Stojanovic, Natalija and Stojnev Ilic, Aleksandra and Weidmann, Vera and Lowitzki, Leon and Wagner, Markus and Ifa, Abdessatar Ben and Arlos, Patrik and Megia, Ana and Vendrell, Joan and Pfitzner, Bjarne and Redondo, Alberto and R{\´i}os Insua, David and Albert, Justin Amadeus and Zhou, Lin and Arnrich, Bert and Szab{\´o}, Ildik{\´o} and Fodor, Szabina and Ternai, Katalin and Bhowmik, Rajarshi and Campero Durand, Gabriel and Shevchenko, Pavlo and Malysheva, Milena and Prymak, Ivan and Saake, Gunter},
  title     = {HPI Future SOC Lab - Proceedings 2019},
  number    = {158},
  editor    = {Meinel, Christoph and Polze, Andreas and Beins, Karsten and Strotmann, Rolf and Seibold, Ulrich and R{\"o}dszus, Kurt and M{\"u}ller, J{\"u}rgen},
  publisher = {Universit{\"a}tsverlag Potsdam},
  address   = {Potsdam},
  isbn      = {978-3-86956-564-4},
  issn      = {1613-5652},
  doi       = {10.25932/publishup-59791},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-597915},
  publisher      = {Universit{\"a}t Potsdam},
  pages     = {xi, 301},
  year      = {2023},
  abstract  = {The "HPI Future SOC Lab" is a cooperation of the Hasso Plattner Institute (HPI) and industry partners. Its mission is to enable and promote exchange and interaction between the research community and the industry partners. The HPI Future SOC Lab provides researchers with free of charge access to a complete infrastructure of state of the art hard and software. This infrastructure includes components, which might be too expensive for an ordinary research environment, such as servers with up to 64 cores and 2 TB main memory. The offerings address researchers particularly from but not limited to the areas of computer science and business information systems. Main areas of research include cloud computing, parallelization, and In-Memory technologies. This technical report presents results of research projects executed in 2019. Selected projects have presented their results on April 9th and November 12th 2019 at the Future SOC Lab Day events.},
  language  = {en}
}