Private Federated Frequency Estimation: Adapting to the Hardness of the Instance

Authors: Jingfeng Wu, Wennan Zhu, Peter Kairouz, Vladimir Braverman

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conclude our work by showing how differential privacy can be added to our algorithm and verifying its superior performance through extensive experiments conducted on large-scale datasets.
Researcher Affiliation Collaboration Jingfeng Wu Johns Hopkins University uuujf@jhu.edu Wennan Zhu Google Research wennanzhu@google.com Peter Kairouz Google Research kairouz@google.com Vladimir Braverman Rice University vb21@rice.edu
Pseudocode Yes Algorithm 1 COUNT SKETCH FOR FEDERATED FREQUENCY ESTIMATION
Open Source Code No The paper does not provide any links to open-source code or explicitly state that the code will be made available.
Open Datasets Yes In the first set of experiments, we simulate a single-round FFE problem with the Gowalla dataset [Cho et al., 2011]. ... In the second set of experiments, we run simulations on the Colossal Clean Crawled Corpus (C4) dataset [Bowman et al., 2020] ... In the third set of experiments, we run simulations on a Twitter dataset Sentiment-140 [Go et al., 2009].
Dataset Splits No The paper describes how problem instances are constructed from datasets (e.g., sampling clients) and parameters for the sketching algorithm are set. However, it does not provide specific train/validation/test dataset splits in the traditional machine learning sense for model training and evaluation.
Hardware Specification No The paper mentions running 'simulations' and 'experiments' but does not specify any hardware details like GPU/CPU models, processors, or memory.
Software Dependencies No The paper describes the algorithms and their theoretical properties but does not specify any software dependencies with version numbers used for implementation or experimentation.
Experiment Setup Yes In the experiments, we fix the confidence parameter to be p = 0.1 and the sketch length to be L = ln(2d/p) ≈ 16. The targeted ℓ∞-error τ is chosen evenly from (10−3, 10−1). ... We set the number of rounds to be M = 10. In each round, n = N/M = 17, 500 clients participate.