FedPFT: One-shot Federated Learning with Foundation Models

Note: If you use this baseline in your work, please remember to cite the original authors of the paper as well as the Flower paper.

Paper: arxiv.org/abs/2402.01862

Authors: Mahdi Beitollahi, Alex Bie, Sobhan Hemati, Leo Maxime Brunswic, Xu Li, Xi Chen, Guojun Zhang.

Abstract: In one-shot federated learning (FL), clients collaboratively train a global model in a single round of communication. Existing approaches for one-shot FL enhance communication efficiency at the expense of diminished accuracy. This paper introduces FedPFT (Federated Learning with Parametric Feature Transfer), a methodology that harnesses the transferability of foundation models to enhance both accuracy and communication efficiency in one-shot FL. The approach involves transferring per-client parametric models (specifically, Gaussian mixtures) of features extracted from foundation models. Subsequently, each parametric model is employed to generate synthetic features for training a classifier head. Experimental results on eight datasets demonstrate that FedPFT enhances the communication-accuracy frontier in both centralized and decentralized FL scenarios, as well as across diverse data-heterogeneity settings such as covariate shift and task shift, with improvements of up to 20.6%. Additionally, FedPFT adheres to the data minimization principle of FL, as clients do not send real features. We demonstrate that sending real features is vulnerable to potent reconstruction attacks. Moreover, we show that FedPFT is amenable to formal privacy guarantees via differential privacy, demonstrating favourable privacy-accuracy tradeoffs.

About this baseline

What’s implemented: The code in this directory replicates the centralized experiments in Parametric Feature Transfer, One-shot Federated Learning with Foundation Models (Beitollahi et al., 2024) for CIFAR-100 and Caltech101 datasets, which proposed the FedPFT algorithm. Concretely, it replicates the results in Section 5.2.

Datasets: CIFAR-100 and Caltech101 from HuggingFace

Hardware Setup: These experiments were run on a desktop machine with 8 CPU threads and Nvidia 4070 with 8GB of VRAM.

Contributors: Mahdi Beitollahi (mahdi.beitollahi@queensu.ca).

Experimental Setup

Task: Image classification

Model: This directory utilizes two pre-trained, frozen models as shown in Table 1 of the paper:

• ResNet50 pre-trained on ImageNet is used for CIFAR-100 dataset(see models/resnet50).

• CLIP, ViT-B/32 pre-trained on web dataset is used for Caltech101 dataset (see models/clip_vit)

Dataset: This baseline includes the CIFAR-100 and Caltech101 datasets via flwr-datasets. By default, it will be partitioned into 50 clients following a Dirichlet distribution with \(\alpha\)=0.1.

Dataset	#classes	#partitions	partitioning method	partition settings
CIFAR-100	100	50	Dirichlet distribution	\(\alpha\)=0.1
Caltech101	101	50	Dirichlet distribution	\(\alpha\)=0.1

Training Hyperparameters: The following table shows the main hyperparameters for this baseline with their default value (i.e. the value used if you run python main.py directly)

Description	Default Value
total clients	50
clients per round	50
number of rounds	1
client resources	{‘num_cpus’: 2.0, ‘num_gpus’: 0.0 }
data partition	distribution with \(\alpha\)=0.1
Number of mixtures	1
Covariance type	spherical
tolerance	1e-12
maximum EM iterations	1e3

Environment Setup

To construct the Python environment, simply run:

# Set directory to use python 3.10 (install with `pyenv install <version>` if you don't have it)
pyenv local 3.10.12

# Tell poetry to use python3.10
poetry env use 3.10.12

# Install
poetry install

Running the Experiments

To run this FedPFT with CIFAR-100 baseline, first ensure you have activated your Poetry environment (execute poetry shell from this directory), then:

python -m fedpft.main # this will run using the default settings in the `conf/config.yaml`

# you can override settings directly from the command line
python -m fedpft.main dataset=Caltech101 model=clip # will set dataset to Caltech101 and the pre-trained model to Clip-ViT/B32

To run using FedAvg:

# this will use a frozen, pre-trained model and train the classifier head
python -m fedpft.main strategy=FedAvg client=FedAvg num_rounds=20 dataset=Caltech101 model=clip num_gpus=0.2

Expected Results

With the following command, we run both FedPFT and FedAvg configurations.

# FedPFT
python -m fedpft.main dataset=CIFAR100 model=resnet50
python -m fedpft.main dataset=Caltech101 model=clip

# FedAvg with pre-trained, frozen models
python -m fedpft.main strategy=fedavg client=fedavg dataset=CIFAR100 model=resnet50 num_rounds=20 strategy.on_fit_config_fn.num_epochs=1 num_gpus=0.5
python -m fedpft.main strategy=fedavg client=fedavg dataset=Caltech101 model=clip num_rounds=20 num_gpus=0.2

The above commands would generate results that you can plot and would look like the plot shown below. This plot was generated using the jupyter notebook in the docs/ directory of this baseline after running the commands above.