Currently, salience-based channel pruning makes continuous breakthroughs in network compression. In the realization, the salience mechanism is used as a metric of channel salience to guide pruning. Therefore, salience-based channel pruning can dynamically adjust the channel width at run-time, which provides a flexible pruning scheme. However, there are two problems emerging: a gating function is often needed to truncate the specific salience entries to zero, which destabilizes the forward propagation; dynamic architecture brings more cost for indexing in inference which bottlenecks the inference speed. In this paper, we propose a Progressive Channel-Shrinking (PCS) method to compress the selected salience entries at run-time instead of roughly approximating them to zero. We also propose a Running Shrinking Policy to provide a testing-static pruning scheme that can reduce the memory access cost for filter indexing. We evaluate our method on ImageNet and CIFAR10 datasets over two prevalent networks: ResNet and VGG, and demonstrate that our PCS outperforms all baselines and achieves state-of-the-art in terms of compression-performance tradeoff. Moreover, we observe a significant and practical acceleration of inference.
Dynamic neural networks can greatly reduce computation redundancy without compromising accuracy by adapting their structures based on the input. In this paper, we explore the robustness of dynamic neural networks against energy-oriented attacks targeted at reducing their efficiency. Specifically, we attack dynamic models with our novel algorithm GradMDM. GradMDM is a technique that adjusts the direction and the magnitude of the gradients to effectively find a small perturbation for each input, that will activate more computational units of dynamic models during inference. We evaluate GradMDM on multiple datasets and dynamic models, where it outperforms previous energy-oriented attack techniques, significantly increasing computation complexity while reducing the perceptibility of the perturbations.
Load forecasting is very essential in the analysis and grid planning of power systems. For this reason, we first propose a household load forecasting method based on federated deep learning and non-intrusive load monitoring (NILM). For all we know, this is the first research on federated learning (FL) in household load forecasting based on NILM. In this method, the integrated power is decomposed into individual device power by non-intrusive load monitoring, and the power of individual appliances is predicted separately using a federated deep learning model. Finally, the predicted power values of individual appliances are aggregated to form the total power prediction. Specifically, by separately predicting the electrical equipment to obtain the predicted power, it avoids the error caused by the strong time dependence in the power signal of a single device. And in the federated deep learning prediction model, the household owners with the power data share the parameters of the local model instead of the local power data, guaranteeing the privacy of the household user data. The case results demonstrate that the proposed approach provides a better prediction effect than the traditional methodology that directly predicts the aggregated signal as a whole. In addition, experiments in various federated learning environments are designed and implemented to validate the validity of this methodology.