Federated Causality Learning with Explainable Adaptive Optimization

Authors: Dezhi Yang, Xintong He, Jun Wang, Guoxian Yu, Carlotta Domeniconi, Jinglin Zhang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on synthetic and real datasets show that Fed Causal can effectively deal with non-independently and identically distributed (non-IID) data and has a superior performance.
Researcher Affiliation Academia Dezhi Yang1, 2, Xintong He3, Jun Wang2*, Guoxian Yu1, 2, Carlotta Domeniconi4, Jinglin Zhang5 1School of Software, Shandong University, Jinan, China 2SDU-NTU Joint Centre for AI Research, Shandong University, Jinan, China 3Department of Mathematics, National University of Singapore, Singapore 4Departiment of Computer Science, George Mason University, VA, USA 5School of Control Science and Engineering, Shandong University, Jinan, China dzyang@mail.sdu.edu.cn, E0966399@u.nus.edu.sg, {kingjun, gxyu,jinglin.zhang}@sdu.edu.cn, carlotta@cs.gmu.edu
Pseudocode Yes Algorithm 1: Fed Causal: Federated Causal Learning
Open Source Code Yes The code of Fed Causal is shared at https://www.sdu-idea.cn/pubDetail?pubId=279.
Open Datasets Yes We evaluate Fed Causal on a protein signaling network based on expression levels of proteins and phospholipid given by (Sachs et al. 2005), which is generally accepted by the biology community.
Dataset Splits No The paper describes generating data or distributing samples to clients (e.g., 'generate 200 samples for each client', 'randomly select n = 7460 samples and evenly distribute them to 10 clients'), but it does not specify explicit training, validation, and test dataset splits (e.g., percentages or counts for each).
Hardware Specification Yes We use the same server (Ubuntu 18.04.5, Intel Xeon Gold 6248R and Nvidia RTX 3090) to perform experiments and report the structural hamming distance (SHD), true positive rate (TPR) and false discovery rate (FDR) of the estimated DAGs, averaged over 10 random runs.
Software Dependencies Yes We use the same server (Ubuntu 18.04.5, Intel Xeon Gold 6248R and Nvidia RTX 3090) to perform experiments and report the structural hamming distance (SHD), true positive rate (TPR) and false discovery rate (FDR) of the estimated DAGs, averaged over 10 random runs.
Experiment Setup No We provide the hyperparameter settings of Fed Causal and other baselines in the supplementary file (Yang et al. 2023).