Enhancing Learning with Label Differential Privacy by Vector Approximation

Authors: Puning Zhao, Jiafei Wu, Zhe Liu, Li Shen, Zhikun Zhang, Rongfei Fan, Le Sun, Qingming Li

ICLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we conduct experiments on both synthesized and real datasets, which validate our theoretical analysis as well as the practical performance of our method. Numerical results on synthesized data validate the theoretical analysis, which shows that the performance of our method only decays slightly with K. Experiments on real datasets also validate the effectiveness of our proposed method.
Researcher Affiliation Academia Puning Zhao, Jiafei Wu, Zhe Liu Zhejiang Lab Hangzhou, Zhejiang, China EMAIL Li Shen Shenzhen Campus of Sun Yat-sen University Shenzhen, Guangdong, China EMAIL Zhikun Zhang Zhejiang University Hangzhou, Zhejiang, China EMAIL Rongfei Fan Beijing Institute of Technology EMAIL Le Sun Nanjing University of Information Science and Technology EMAIL Qingming Li Zhejiang University EMAIL
Pseudocode No The paper describes methods and concepts in paragraph text and uses a figure (Figure 1) to illustrate a comparison, but it does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not contain an explicit statement about releasing source code or provide a link to a code repository.
Open Datasets Yes Now we evaluate our new method on standard benchmark datasets that have been widely used in previous works on differentially private machine learning, including MNIST (Le Cun, 1998), Fashion MNIST (Xiao et al., 2017) CIFAR-10 and CIFAR-100(Krizhevsky et al., 2009).
Dataset Splits No The paper mentions using standard benchmark datasets (MNIST, Fashion MNIST, CIFAR-10, CIFAR-100) but does not explicitly state the training/test/validation split ratios, sample counts, or specific methodology for creating these splits. It implies the use of standard splits for these datasets.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments.
Software Dependencies No The paper mentions using 'Adam optimizer' and types of neural networks ('simple convolutional neural network', 'ResNet-18'), but it does not specify any software names with version numbers (e.g., Python version, PyTorch/TensorFlow versions, or specific library versions).
Experiment Setup Yes For the MNIST and Fashion MNIST datasets, we use a simple convolutional neural network composed of two convolution and pooling layers with dropout rate 0.5... In our experiments, we set the batch size to be 400, and use the Adam optimizer with learning rate 0.001.