Formal Logic Enabled Personalized Federated Learning through Property Inference

Authors: Ziyan An, Taylor T. Johnson, Meiyi Ma

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate the proposed method on two tasks: a real-world traffic volume prediction task consisting of sensory data from fifteen states and a smart city multi-task prediction utilizing synthetic data. The evaluation results exhibit clear improvements, with performance accuracy improved by up to 54% across all sequential prediction models.
Researcher Affiliation Academia Ziyan An, Taylor T. Johnson, Meiyi Ma Department of Computer Science, Vanderbilt University, Nashville, TN, USA {ziyan.an, taylor.johnson, meiyi.ma}@vanderbilt.edu
Pseudocode Yes Algorithm 1: CLUSTER ID: Cluster identity mapping; Algorithm 2: Fed STL: Client federation and update
Open Source Code Yes Code implementation is available at https://github.com/AICPSLab/Fed STL.git.
Open Datasets Yes We obtain a publicly available dataset from the Federal Highway Administration (FHWA 2016) and preprocess hourly traffic volume from 15 states. ... we create a simulated dataset using SUMO (Simulation of Urban MObility) (Krajzewicz et al. 2002), a large-scale open-source road traffic simulator.
Dataset Splits No The paper does not provide specific details on training, validation, or test dataset splits (e.g., percentages, sample counts, or k-fold cross-validation setup).
Hardware Specification Yes The experiments were conducted on a machine equipped with an Intel Core i9-10850K CPU and an NVIDIA GeForce RTX 3070 GPU.
Software Dependencies No The paper mentions 'The operating system used was Ubuntu 18.04.' but does not list specific versions for other key software components like machine learning frameworks (e.g., PyTorch, TensorFlow) or libraries.
Experiment Setup Yes During each round of FL communication, we randomly select 10% of the client devices to participate. For all the conducted experiments and algorithms, we use SGD with consistent learning rates and a batch size of 64. ... We set the number of local epochs to 10 for Fed Avg, Fed Prox, Fed Rep (with 8 head epochs), Ditto, and IFCA. Additionally, for Fed STL, we employ 6 local epochs and 4 cluster training epochs.