FedMM

This is the official PyTorch implementation for the paper FedMM: Federated Collaborative Signal Quantization for Multi-Market CTR Prediction.

📢 "FedMM" has been accepted by SIGIR 2026!

Framework

Pipeline

Step 1: Data Processing

# data preprocess
python data/preprocess/0_item_feature.py
python data/preprocess/1_gen_label_multi_countrys.py
# feature map
python data/preprocess/2_feat_map.py

# split data for cf model
python data/preprocess/3_cf_split.py
# split data for ctr model
python data/preprocess/3_ctr_split.py

Step 2: Get CF Embeddings

# Use Canada (ca) as an example
python trainer/cf_trainer.py --country ca

Step 3: FedMM

python trainer/fedmm_main.py --countrys ca de fr it

Step 4: Enhance CTR Prediction

python trainer/ctr_trainer.py --country ca