联邦学习：去中心化数据下的深度网络优化 - IT评测·应用市场-qidao123.com

import random
import numpy as np
def compute_loss(model, data):
"""计算模型在给定数据上的损失值（这里使用简单的均方误差）"""
# 假设数据的最后一列是标签
X = data[:, :-1]
y = data[:, -1]
predictions = np.dot(X, model)
loss = np.mean((predictions - y) ** 2)
return loss
def initialize_model():
"""初始化全局模型参数"""
return np.random.rand(9) # 改为9维，因为最后一列是标签
def client_update(model, data, epochs, lr):
"""客户端本地训练"""
losses = []
for _ in range(epochs):
gradient = compute_gradient(model, data)
model -= lr * gradient
loss = compute_loss(model, data)
losses.append(loss)
return model, np.mean(losses)
def compute_gradient(model, data):
"""计算梯度（使用均方误差的梯度）"""
X = data[:, :-1]
y = data[:, -1]
predictions = np.dot(X, model)
gradient = -2 * np.dot(X.T, (y - predictions)) / len(data)
return gradient
def aggregate_models(client_models, num_samples):
"""聚合客户端模型参数（加权平均）"""
total_samples = sum(num_samples)
weights = [n / total_samples for n in num_samples]
new_global_model = np.zeros_like(client_models[0])
for model, weight in zip(client_models, weights):
new_global_model += model * weight
return new_global_model
def federated_training(num_rounds, num_clients, fraction, local_epochs, lr):
"""联邦训练过程"""
global_model = initialize_model()
# 生成模拟数据：每个客户端100条数据，每条数据9个特征和1个标签
client_data = {
i: np.concatenate([
np.random.rand(100, 9), # 特征
np.random.rand(100, 1) # 标签
], axis=1)
for i in range(num_clients)
}
global_losses = []
for round in range(num_rounds):
selected_clients = random.sample(range(num_clients), max(1, int(fraction * num_clients)))
client_models = []
client_losses = []
num_samples = []
for client in selected_clients:
local_model = global_model.copy()
updated_model, local_loss = client_update(local_model, client_data[client], local_epochs, lr)
client_models.append(updated_model)
client_losses.append(local_loss)
num_samples.append(len(client_data[client]))
# 计算这一轮的平均损失
avg_loss = np.mean(client_losses)
global_losses.append(avg_loss)
global_model = aggregate_models(client_models, num_samples)
print(f"Round {round+1}: Average Loss = {avg_loss:.6f}")
print("\n训练完成！")
print(f"初始损失值: {global_losses[0]:.6f}")
print(f"最终损失值: {global_losses[-1]:.6f}")
print(f"损失下降率: {((global_losses[0] - global_losses[-1]) / global_losses[0] * 100):.2f}%")
return global_model, global_losses
# 运行联邦学习
final_model, losses = federated_training(num_rounds=10, num_clients=5, fraction=0.6, local_epochs=5, lr=0.1)

复制代码