Understanding treatment effect heterogeneity is vital for scientific and policy research.

We provide our source codes to ensure the reproducibility of our experimental results. Below we summarize several critical aspects w.r .tthe The datasets we used are all publicly accessible, which is introduced in Appendix E.1. For long-tailed subsets, we strictly follows previous work [29] on CIFAR-100-L T to avoid the bias attribute to the sampling randomness. On ImageNet-L T and Places-L T, we employ the widely-used data split first introduced in [44]. All the experiments are conducted on NVIDIA GeForce RTX 3090 with Python 3.7 and Pytorch 1.7.

In Table 1, we present the results on the Federated EMNIST (FEMNIST)16 dataset which isone of therealisticfederated learning datasets in the literature (see the paper by Caldas etal. In this experiments, for IFCA and one-shot clustering algorithm, we share the representation layers among all the20 models, but the last layers for different models are trained based on clustering. As we can see, the results of IFCA21 areonparwith theone-shot clustering algorithm. In our experiments, we observe that our algorithm is robust to the choice of27 numberofclusters.

In this paper, we initiate the study of Euclidean clustering with Distance-based privacy. Distance-based privacy is motivated by the fact that it is often only needed to protect the privacy of exact, rather than approximate, locations.

However,ithas attracted less attention when the objects are difficult to represent in astandard way,for example cars or food.