ABSTRACT
In this paper, we present a new AI (Artificial Intelligence) edge platform, called "MiniDeep", which provides a standalone deep learning platform based on the cloud-edge architecture. This AI-Edge platform provides developers with a whole deep learning development environment to set up their deep learning life cycle processes, such as model training, model evaluation, model deployment, model inference, ground truth collecting, data pre-processing, and training data management. To the best of our knowledge, such a whole deep learning development environment has not been built before. MiniDeep uses Amazon Web Services (AWS) as the backend platform of a deep learning tuning management model. In the edge device, the OpenVino enables deep learning inference acceleration at the edge. To perform a deep learning life cycle job, MiniDeep proposes a mini deep life cycle (MDLC) system which is composed of several microservices from the cloud to the edge. MiniDeep provides Train Job Creator (TJC) for training dataset management and the models' training schedule and Model Packager (MP) for model package management. All of them are based on several AWS cloud services. On the edge device, MiniDeep provides Inference Handler (IH) to handle deep learning inference by hosting RESTful API (Application Programming Interface) requests/responses from the end device. Data Provider (DP) is responsible for ground truth collection and dataset synchronization for the cloud. With the deep learning ability, this paper uses the MiniDeep platform to implement a recommendation system for AI-QSR (Quick Service Restaurant) KIOSK (interactive kiosk) application. AI-QSR uses the MiniDeep platform to train an LSTM (Long Short-Term Memory)-based recommendation system. The LSTM-based recommendation system converts KIOSK UI (User Interface) flow to the flow sequence and performs sequential recommendations with food suggestions. At the end of this paper, the efficiency of the proposed MiniDeep is verified through real experiments. The experiment results have demonstrated that the proposed LSTM-based scheme performs better than the rule-based scheme in terms of purchase hit accuracy, categorical cross-entropy, precision, recall, and F1 score.
Subject(s)
Artificial Intelligence , Deep Learning , SoftwareABSTRACT
Indoor localization is an important issue for indoor location-based services. As opposed to the other indoor localization approaches, the radio frequency (RF) based approaches are low-energy solutions with simple implementation. The kernel learning has been used for the RF-based indoor localization in 2D environment. However, the kernel learning has not been used in 3D environment. Hence, this paper proposes a multi-kernel learning scheme for 3D indoor localization. Based on the signals collected in the area of interest, the WiFi signals with better quality and closer to the user are selected so as to reduce the multipath effect and the external interference. Through the construction of multi-kernel, the localization accuracy can be improved as opposed to the localization based on the single kernel. We build multiple kernels to get the user's location by collecting wireless received signal strengths (RSS) and signal-to-noise ratios (SNR). The kernel learning maps data to high dimension space and uses the optimization process to find the surface where the data are mapped. By multi-kernel training, the surface is fine-tuned and eventually converges to form the location database during the mapping process. The proposed localization scheme is verified by the real RSS and SNR collected from multiple wireless access points (AP) in a building. The experimental results verify that the proposed multi-kernel learning scheme performs better than the multi-DNN scheme and the existing kernel-based localization schemes in terms of localization accuracy and error in 3D indoor environment.
ABSTRACT
During the training phase of the supervised learning, it is not feasible to collect all the datasets of labelled data in an outdoor environment for the localization problem. The semi-supervised transfer learning is consequently used to pre-train a small number of labelled data from the source domain to generate a kernel knowledge for the target domain. The kernel knowledge is transferred to a target domain to transfer some unlabelled data into the virtual labelled data. In this paper, we have proposed a new outdoor localization scheme using a semi-supervised transfer learning for LoRaWANs. In the proposed localization algorithm, a grid segmentation concept is proposed so as to generate a number of virtual labelled data through learning by constructing the relationship of labelled and unlabelled data. The labelled-unlabelled data relationship is repeatedly fine-tuned by correctly adding some more virtual labelled data. Basically, the more the virtual labelled data are added, the higher the location accuracy will be obtained. In the real implementation, three types of signal features, RSSI, SNR, and timestamps, are used for training to improve the location accuracy. The experimental results illustrate that the proposed scheme can improve the location accuracy and reduce the localization error as opposed to the existing outdoor localization schemes.