No abstract

With the development of the mobile systems, we gain a lot of benefits and convenience by leveraging mobile devices; at the same time, the information gathered by smartphones, such as location and environment, is also valuable for business to provide more intelligent services for customers. More and more machine learning methods have been used in the field of mobile information systems to study user behavior and classify usage patterns, especially convolutional neural network. With the increasing of model training parameters and data scale, the traditional single machine training method cannot meet the requirements of time complexity in practical application scenarios. The current training framework often uses simple data parallel or model parallel method to speed up the training process, which is why heterogeneous computing resources have not been fully utilized. To solve these problems, our paper proposes a delay synchronization convolutional neural network parallel strategy, which leverages the heterogeneous system. The strategy is based on both synchronous parallel and asynchronous parallel approaches; the model training process can reduce the dependence on the heterogeneous architecture in the premise of ensuring the model convergence, so the convolution neural network framework is more adaptive to different heterogeneous system environments. The experimental results show that the proposed delay synchronization strategy can achieve at least three times the speedup compared to the traditional data parallelism.

In order to utilize the distributed characteristic of sensors, distributed machine learning has become the mainstream approach, but the different computing capability of sensors and network delays greatly influence the accuracy and the convergence rate of the machine learning model. Our paper describes a reasonable parameter communication optimization strategy to balance the training overhead and the communication overhead. We extend the fault tolerance of iterative-convergent machine learning algorithms and propose the Dynamic Finite Fault Tolerance (DFFT). Based on the DFFT, we implement a parameter communication optimization strategy for distributed machine learning, named Dynamic Synchronous Parallel Strategy (DSP), which uses the performance monitoring model to dynamically adjust the parameter synchronization strategy between worker nodes and the Parameter Server (PS). This strategy makes full use of the computing power of each sensor, ensures the accuracy of the machine learning model, and avoids the situation that the model training is disturbed by any tasks unrelated to the sensors.