Edge computing enables low-latency and privacy-preserving DNN inference, yet heterogeneous and dynamically changing device resources make it difficult to satisfy real-time constraints. In this paper, we present AdaEnsemble, an adaptive and collaborative ensemble inference framework that integrates Dynamic DNNs with deadline-aware scheduling. The system profiles accuracy and latency offline and selects both model widths and participating devices at runtime to maximize accuracy under a given deadline. Experiments on heterogeneous edge devices show that AdaEnsemble adapts effectively to different latency requirements and consistently outperforms the state-of-the-art.

Distributed inference is a popular approach for efficient DNN inference at the edge. However, traditional Static and Dynamic DNNs are not distribution-friendly, causing system reliability and adaptability issues. In this paper, we introduce Fluid Dynamic DNNs (Fluid DyDNNs), tailored for distributed inference. Distinct from Static and Dynamic DNNs, Fluid DyDNNs utilize a novel nested incremental training algorithm to enable independent and combined operation of its sub-networks, enhancing system reliability and adaptability. Evaluation on embedded Arm CPUs with a DNN model and the MNIST dataset, shows that in scenarios of single device failure, Fluid Dy DNNs ensure continued inference, whereas Static and Dynamic DNNs fail. When devices are fully operational, Fluid DyDNNs can operate in either a High-Accuracy mode and achieve comparable accuracy with Static DNNs, or in a High-Throughput mode and achieve 2.5x and 2x throughput compared with Static and Dynamic DNNs, respectively.

Dynamic Deep Neural Networks (DNNs) can achieve faster execution and less computationally intensive inference by spending fewer resources on easy to recognise or less informative parts of an input. They make data-dependent decisions, which strategically deactivate a model's components, e.g. layers, channels or sub-networks. However, dynamic DNNs have only been explored and applied on conventional computing systems ($\text{CPU} +\text{GPU}$)) and programmed with libraries designed for static networks, limiting their effects. In this paper, we propose and explore two approaches for efficiently realising the sub-networks that make these decisions on FPGAs. A pipeline approach targets the use of the existing hardware to execute the sub-network, while a parallel approach uses dedicated circuitry for it. We explore the performance of each using the BranchyNet early exit approach on LeNet-5, and evaluate on a Xilinx ZCU106. The pipeline approach is 36% faster than a desktop CPU. It consumes 0.51 mJ per inference, 16x lower than a non-dynamic network on the same platform and 8x lower than an Nvidia Jetson Xavier NX. The parallel approach executes 17% faster than the pipeline approach when on dynamic inference no early exits are taken, but incurs an increase in energy consumption of 28%.

With the rapid industrialization and urbanization in China, the great increases in anthropogenic emissions during the last decades have caused serious air pollution problems greatly influencing public health [1-2]. The Beijing-Tianjin-Hebei (BTH) region, located in central-eastern China, is not only one of China's most economically developed and industrialized regions, but is the area that most frequently experiences haze episodes. In this study, the characteristics of spatial and temporal distribution and variation trends of aerosol optical depth (AOD) on 550 nm, SO2 and NO2 column density during 2002-2013 over BTH region were analyzed by using MODIS, OMI, GOME-2 and SCIAMACHY satellite data [3-5]. With the Geographically Weighted Regression (GWR) model, the driving factors including motor vehicle ownership, electricity output, population and the construction area from the city statistical yearbooks are preliminary analyzed.