论文部分内容阅读
为降低无线传感器网络中核学习机训练时的数据通信代价和节点计算代价,研究了基于筛选机制的L1正则化核学习机分布式训练方法。提出了一种节点局部训练样本筛选机制,各节点利用筛选出的训练样本,在节点模型对本地训练样本的预测值与邻居节点间局部最优模型对本地训练样本预测值相一致的约束下,利用增广拉格朗日乘子法求解L1正则化核学习机分布式优化问题,利用交替方向乘子法求解节点本地的L1正则化核学习机的稀疏模型;仅依靠相邻节点间传输稀疏模型的协作方式,进一步优化节点局部模型,直至各节点模型收敛。基于此方法,提出了基于筛选机制的L1正则化核最小平方误差学习机的分布式训练算法。仿真实验验证了该算法在模型预测正确率、模型稀疏率、数据传输量和参与模型训练样本量上的有效性和优势。
In order to reduce the data communication cost and node calculation cost of nuclear learning machine training in wireless sensor networks, a distributed training method of L1 regularized kernel learning machine based on screening mechanism is studied. This paper proposes a screening mechanism of local training samples for nodes. Using the selected training samples, each node is constrained by the prediction of local training samples by the node model and the prediction of local training samples by the local optimal model of neighboring nodes. The augmented Lagrange multiplier method is used to solve the distributed optimization problem of L1 regularized nuclear learning machine. The alternating direction multiplier method is used to solve the sparse model of L1 regularized nuclear learning machine local to the node. Relying on sparse transmission between adjacent nodes, Model collaboration mode, and further optimize the local node model until each node model convergence. Based on this method, a distributed training algorithm for L1 regularized kernel least square error learning machine based on screening mechanism is proposed. Simulation results show the effectiveness and advantage of the proposed algorithm in predicting the correctness of the model, the sparseness of the model, the amount of data transferred and the training samples involved in the model.