Refine
Year of publication
- 2021 (15) (remove)
Document type
Is part of the Bibliography
- Yes (15)
Keywords
- IoT (3)
- Blockchain (2)
- Machine learning (2)
- Resource discovery (2)
- Security (2)
- Artificial neural network (1)
- Assessment (1)
- Audit (1)
- Chatbot (1)
- Cloud (1)
Nowadays, machine learning projects have become more and more relevant to various real-world use cases. The success of complex Neural Network models depends upon many factors, as the requirement for structured and machine learning-centric project development management arises. Due to the multitude of tools available for different operational phases, responsibilities and requirements become more and more unclear. In this work, Machine Learning Operations (MLOps) technologies and tools for every part of the overall project pipeline, as well as involved roles, are examined and clearly defined. With the focus on the inter-connectivity of specific tools and comparison by well-selected requirements of MLOps, model performance, input data, and system quality metrics are briefly discussed. By identifying aspects of machine learning, which can be reused from project to project, open-source tools which help in specific parts of the pipeline, and possible combinations, an overview of support in MLOps is given. Deep learning has revolutionized the field of Image processing, and building an automated machine learning workflow for object detection is of great interest for many organizations. For this, a simple MLOps workflow for object detection with images is portrayed.
While the number of devices connected together as the Internet of Things (IoT) is growing, the demand for an efficient and secure model of resource discovery in IoT is increasing. An efficient resource discovery model distributes the registration and discovery workload among many nodes and allow the resources to be discovered based on their attributes. In most cases this discovery ability should be restricted to a number of clients based on their attributes, otherwise, any client in the system can discover any registered resource. In a binary discovery policy, any client with the shared secret key can discover and decrypt the address data of a registered resource regardless of the attributes of the client. In this paper we propose Attred, a decentralized resource discovery model using the Region-based Distributed Hash Table (RDHT) that allows secure and location-aware discovery of the resources in IoT network. Using Attribute Based Encryption (ABE) and based on predefined discovery policies by the resources, Attred allows clients only by their inherent attributes, to discover the resources in the network. Attred distributes the workload of key generations and resource registration and reduces the risk of central authority management. In addition, some of the heavy computations in our proposed model can be securely distributed using secret sharing that allows a more efficient resource registration, without affecting the required security properties. The performance analysis results showed that the distributed computation can significantly reduce the computation cost while maintaining the functionality. The performance and security analysis results also showed that our model can efficiently provide the required security properties of discovery correctness, soundness, resource privacy and client privacy.
In Industry 4.0 machine learning approaches are a state-of-the art for predictive maintenance, machine condition monitoring, and others. Distributed decision trees are one of the learning algorithms for such applications. A new approach of node based parallelization for the construction is presented and allows to classify data through a network of nodes. Attacks on the nodes are discussed based on different attack scenarios and attack classifications are presented. A thorough analysis of protection measurements is given, such that classification is not maliciously modified by an attacker. Different countermeasures are proposed and analyzed. A quorum-based system allows for a good balance between computational overhead and robustness of the algorithm.
Distributed machine learning algorithms that employ Deep Neural Networks (DNNs) are widely used in Industry 4.0 applications, such as smart manufacturing. The layers of a DNN can be mapped onto different nodes located in the cloud, edge and shop floor for preserving privacy. The quality of the data that is fed into and processed through the DNN is of utmost importance for critical tasks, such as inspection and quality control. Distributed Data Validation Networks (DDVNs) are used to validate the quality of the data. However, they are prone to single points of failure when an attack occurs. This paper proposes QUDOS, an approach that enhances the security of a distributed DNN that is supported by DDVNs using quorums. The proposed approach allows individual nodes that are corrupted due to an attack to be detected or excluded when the DNN produces an output. Metrics such as corruption factor and success probability of an attack are considered for evaluating the security aspects of DNNs. A simulation study demonstrates that if the number of corrupted nodes is less than a given threshold for decision-making in a quorum, the QUDOS approach always prevents attacks. Furthermore, the study shows that increasing the size of the quorum has a better impact on security than increasing the number of layers. One merit of QUDOS is that it enhances the security of DNNs without requiring any modifications to the algorithm and can therefore be applied to other classes of problems.
Training of neural networks requires often high computational power and large memory on Graphics Processing Unit (GPU) hardware. Many cloud providers such as Amazon, Azure, Google, Siemens, etc, provide such infrastructure. However, should one choose a cloud infrastructure or an on premise system for a neural network application, how can these systems be compared with one another? This paper investigates seven prominent Machine Learning benchmarks, which are MLPerf, DAWNBench, DeepBench, DLBS, TBD, AIBench, and ADABench. The recent popularity and widespread use of Deep Learning in various applications have created a need for benchmarking in this field. This paper shows that these application domains need slightly different resources and argue that there is no standard benchmark suite available that addresses these different application needs. We compare these benchmarks and summarize benchmarkrelated datasets, domains, and metrics. Finally, a concept of an ideal benchmark is sketched.
Potentials of Semantic Image Segmentation Using Visual Attention Networks for People with Dementia
(2021)
Due to the increasing number of dementia patients, it is time to include the care sector in digitization as well. Digital media, for example, can be used on tablets in memory care and have considerable potential for reminiscence therapy for people with dementia. The time consuming assembly of digital media content has to be automated for the caretakers.
This work analyzes the potentials of semantic image segmentation with Visual Attention Networks for reminiscence therapy sessions. These approaches enable the selection of digital images to satisfy the patients individual experience and biographically. A detailed comparison of various Visual Attention Networks evaluated by the BLEU score is shown. The most promising networks for semantic image segmentation are VGG16 and VGG19.
In edge/fog computing infrastructures, the resources and services are offloaded to the edge and computations are distributed among different nodes instead of transmitting them to a centralized entity. Distributed Hash Table (DHT) systems provide a solution to organizing and distributing the computations and storage without involving a trusted third party. However, the physical locations of nodes are not considered during the creation of the overlay which causes some efficiency issues. In this paper, Locality aware Distributed Addressing (LADA) model is proposed that can be adopted in distributed infrastructures to create an overlay that considers the physical locations of participating nodes. LADA aims to address the efficiency issues during the store and lookup processes in DHT overlay. Additionally, it addresses the privacy issue in similar proposals and removes any possible set of fixed entities. Our studies showed that the proposed model is efficient, robust and is able to protect the privacy of the locations of the participating nodes.