Search

An increasing number of artificial intelligence (AI) applications involve the execution of deep neural networks (DNNs) on edge devices. Many practical reasons motivate the need to update the DNN model on the edge device post-deployment, such as refining the model, concept drift, or outright change in the learning task. In this paper, we consider the scenario where retraining can be done on the server side based on a copy of the DNN model, with only the necessary data transmitted to the edge to update the deployed model. However, due to bandwidth constraints, we want to minimise the transmission required to achieve the update. We develop a simple approach based on matrix factorisation to compress the model update -- this differs from compressing the model itself. The key idea is to preserve existing knowledge in the current model and optimise only small additional parameters for the update which can be used to reconstitute the model on the edge. We compared our method to similar techniques used in federated learning; our method usually requires less than half of the update size of existing methods to achieve the same accuracy., Comment: CVPR 2022 Mobile AI Workshop

Civil structures are on the verge of changing which leads energy dissipation capacity to decline. Structural Health Monitoring (SHM) as a process in order to implement a damage detection strategy and assess the condition of structure plays a key role in structural reliability. Earthquake is a recognized factor in variation of structures condition, inasmuch as inelastic behavior of a building subjected to design level earthquakes is plausible. In this study Hilbert Huang Transformation (HHT) is superseded by Ensemble Empirical Mode decomposition (EEMD) and Hilbert Transform (HT) together. Albeit analogous, EEMD brings more appropriate Intrinsic Mode Functions (IMFs) than Empirical Mode Decomposition (EMD). IMFs are employed to assess first mode frequency and mode shape. Afterward, Artificial Neural Networks (ANN) is applied to predict story acceleration based on acceleration of structure during previous moments. ANN functions precisely. Therefore, any congruency between predicted and measured acceleration provides onset of damage. Then another ANN method is applied to estimate stiffness matrix. Though first mode shape and frequency is calculated in advance, it essentially requires an inverse problem to be solved in order to find stiffness matrix. This task is done by ANN. In other words, these two ANN methods are exercised to forecast location and measure severity of damage respectively. This algorithm is implemented on one nonlinear moment-resisting steel frame and the results are acceptable.