Task-Agnostic Privacy-Preserving Representation Learning for Federated Learning against Attribute Inference Attacks

Authors: Caridad Arroyo Arevalo, Sayedeh Leila Noorbakhsh, Yun Dong, Yuan Hong, Binghui Wang

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive results on multiple datasets and applications validate the effectiveness of TAPPFL to protect data privacy, maintain the FL utility, and be efficient as well. Experimental results also show that TAPPFL outperforms the existing defenses.
Researcher Affiliation Academia 1Illinois Institute or Technology 2Benedictine University 3University of Connecticut
Pseudocode Yes Algorithm 1 in the full version details the TAPPFL training process.
Open Source Code Yes The proofs are in the full version: https://github.com/TAPPFL.
Open Datasets Yes We evaluate our TAPPFL using three datasets from different applications. CIFAR-10 (Krizhevsky 2009) is an image dataset... For the Loans dataset (Hardt, Price, and Srebro 2016)... For the Adult income dataset (Becker and Kohavi 1996).
Dataset Splits No The paper mentions 'training/testing sets' but does not explicitly specify a validation set or its split percentage/count.
Hardware Specification Yes We use the Chameleon Cloud platform offered by the NSF (Keahey et al. 2020) (Cent OS7-CUDA 11 with Nvidia Rtx 6000).
Software Dependencies Yes We use the Chameleon Cloud platform offered by the NSF (Keahey et al. 2020) (Cent OS7-CUDA 11 with Nvidia Rtx 6000). ... The TAPPFL algorithm is implemented in Py Torch.
Experiment Setup Yes In each device, we train the three parameterized neural networks via the Stochastic Gradient Descent (SGD) algorithm, where we set the local batch size to be 10 and use 10 local epochs, and the learning rate in SGD is 0.01. ... The number of global rounds is set to be 20.