VFL for LR

Vertical Federated Learning Implementation for Logistic regression (The Simplest Version)

graph LR;
    Client1 ==Send embedding data==>Server;
    Server ==Send grads w.r.t. embedding data==> Client1;
    Client1 ==Update the local model==> Client1;
    Server ==Calculate grads and update the global model==> Server;
    Client2
    Client...

Notice: Without Regard To Privacy Preserving (which means that use DP, FE, FHE, and so forth to protect embeding data or grads in transfering process)

Prerequisites

• Python, NumPy, libsvm-official

Run Example

$ bash ./data/adult/get_data.sh
$ python3 example.py

Example Result

$ python3 example.py
[*info] Current Test Loss 9830.097444 Current Test Acc: 0.429904
[*info] Epoch 0 Batch 0 Current Train Loss: 1.034175
[*info] Current Test Loss 9830.097444 Current Test Acc: 0.429904
[*info] Epoch 0 Batch 3 Current Train Loss: 0.723213
[*info] Current Test Loss 8148.651797 Current Test Acc: 0.516067
[*info] Epoch 0 Batch 6 Current Train Loss: 0.537807
[*info] Current Test Loss 7347.467367 Current Test Acc: 0.589458
[*info] Epoch 0 Batch 9 Current Train Loss: 0.386673
[*info] Current Test Loss 7006.886287 Current Test Acc: 0.642879
...
[*info] Current Test Loss 8441.881878 Current Test Acc: 0.759453
[*info] Epoch 4 Batch 3 Current Train Loss: 0.076631
[*info] Current Test Loss 8582.104312 Current Test Acc: 0.760061
[*info] Epoch 4 Batch 6 Current Train Loss: 0.069134
[*info] Current Test Loss 8713.062722 Current Test Acc: 0.760365
[*info] Epoch 4 Batch 9 Current Train Loss: 0.065743
[*info] Current Test Loss 8842.630094 Current Test Acc: 0.760973

Reference

[1] VAFL: a Method of Vertical Asynchronous Federated Learning

[2] VerFedLogistic.jl