Domain adaptation in fault diagnosis can efficiently handle different data distributions by co-training source and target domain data. However, the source domain data may not be accessible due to privacy or memory issues. An effective solution is to perform unsupervised parameter tuning of the source model using unlabeled target domain data. The focus of this article is to perform unsupervised domain adaptation using the knowledge of multiple pretrained source models in Industrial Internet of Things (IIoT) without accessing source data. Almost all of the current researches in this setting requires tuning the entire backbone network, increasing computing costs as model size or source domains grow. To this end, this article proposes an efficient multidomain knowledge fusion adaptation (EMDKFA) method, applied to source free cross-domain fault diagnosis. Unlike previous methods, this method greatly reduces the number of trainable parameters and saves computational overhead via low-rank reparameterization strategy. To better fuse and transfer multiple pretrained models, we propose an unsupervised domain weight initialization method based on nuclear norm, initially focusing credible source models. Furthermore, the multidomain weighted entropy minimization penalty and noisy label learning are designed to promote correct allocation of target samples. Comprehensive cross-condition and cross-device experiments demonstrate the method’s effectiveness.

Convolutional Neural Networks (CNNs) have been applied to camera relocalization, which is to infer the pose of the camera given a single monocular image. However, there are still many open problems for camera relocalization with CNNs. We delve into the CNNs for camera relocalization. First, a variant of Euler angles named Euler6 is proposed to represent orientation. Then a data augmentation method named pose synthesis is designed to reduce spsarsity of poses in the whole pose space to cope with overfitting in training. Third, a multi-task CNN named BranchNet is proposed to deal with the complex coupling of orientation and translation. The network consists of several shared convolutional layers and splits into two branches which predict orientation and translation, respectively. Experiments on the 7Scenes dataset show that incorporating these techniques one by one into an existing model PoseNet always leads to better results. Together these techniques reduce the orientation error by 15.9% and the translation error by 38.3% compared to the state-of-the-art model Bayesian PoseNet. We implement BranchNet on an Intel NUC mobile platform and reach a speed of 43 fps, which meets the real-time requirement of many robotic applications.

Early stage power estimation is critical for SoC architecture exploration and validation in modern VLSI design, but real-time, long time interval and accurate estimation is still challenging for system-level estimation and software/hardware tuning. This work proposes a model abstraction approach for real-time power estimation in the manner of machine learning. The singular value decomposition (SVD) technique is exploited to abstract the principle components of relationship between register toggling profile and accurate power waveform. The abstracted power model is automatically instrumented to RTL implementation and synthesized into FPGA platform for real-time power estimation by instrumenting the register toggling profile. The prototype implementation on three IP cores predicts the cycle-by-cycle power dissipation within 5% accuracy loss compared with a commercial power estimation tool.

In this paper, we propose a novel simulation algorithm for large scale structured power grid networks. The new method formulates the traditional linear system as a special two-dimension Poisson equation and solves it using an analytical expressions based on FFT technique. The computation complexity of the new algorithm is O(NlgN), which is much smaller than the traditional solver's complexity O(N1.5) for sparse matrices, such as the SuperLU solver and the PCG solver. Also, due to the special formulation, graphic process unit (GPU) can be explored to further speed up the algorithm. Experimental results show that the new algorithm is stable and can achieve 100X speed up on GPU over the widely used SuperLU solver with very little memory footprint.