Self-describing and Data Propagation Model for Data Distribution Service

Abstract

Abstract.To realize real-time information sharing in generic platforms, it is especially important to support dynamic message structure changes. For the case of IDL, it is necessary to rewrite applications to change data sample structures. In this paper, we propose a dynamic reconfiguration scheme of data sample structures for DDS. Instead of using IDL, which is the static data sample structure model of DDS, we use a self describing model using data sample schema, as a dynamic data sample structure model to support dynamic reconfiguration of data sample structures. We also propose a data propagation model to provide data persistency in distributed environments. We guarantee persistency by transferring data samples through relay nodes to the receiving nodes, which have not participated in the data distribution network at the data sample distribution time. The proposed schemes can be utilized to support data sample structure changes during operation time and to provide data persistency in various environments, such as real-time enterprise environments and connection-less internet environments.

Efficient data-intensive event-driven interaction in SOA

“…Data-centric pub/sub has been addressed by the Data Distribution Service (DDS) OMG standard [14] and by some related research papers [16,12]. The purpose of the specification is to provide a common application level interface that defines the data-distribution service.…”

Improving image classification robustness using self‐supervision

“…While labels generally improve the classification accuracy, in case of severe imbalance, they also introduce a bias resulting in changed decision boundaries, as Yang and Xu (2020) point out. Imbalance thus still introduces a problem for learning the classifier in a supervised downstream task, but this is at least mitigated by the pretext task (Lee et al, 2019). Self‐supervised learning can help to leverage imbalance due to the pretext task which uses balanced pseudo‐labels instead, consequently the method can outperform standard supervised approaches (Lee et al, 2019; Yang & Xu, 2020).…”

“…In many application problems, the training data exhibit long‐tailed distributions including extremely imbalanced classes (Mazurowski et al, 2008; Murphey et al, 2004). Labels of imbalanced data sets induce a bias, and majority classes can alter the decision boundaries (Lee et al, 2019). During the pretext task, imbalanced labels are ignored, learning on the basis of much more balanced self‐supervised signals, also resulting in a better downstream classification (Yang & Xu, 2020).…”

Communication Within Multi-FSM Based Robotic Systems

“…The use of data-centric communication similar to blackboards has transcended AI and found its way into the wider software engineering community. The Data Distribution Service (DDS) is a standard formulated by the Object Management Group (OMG), implementing Data-Centric Publish/Subscribe (DCPS) communication model [40,58]. The main purpose of DDS is to assure dependable communication in dynamic, networked environments between distributed producers and consumer fulfilling real-time constraints.…”

“…The implementation of the distributed version of the whiteboard uses efficient, fault-tolerant communication for distributed systems as presented in [11,51,57] [52,53]. Compared to the DCPS [58], the whiteboard does not offer inherent mechanisms for distributed reconfiguration, but offers a statically defined data classes and O(1), fail-silent communication within bounded real-time constraints [31]. A more detailed discussion of the networked version of the whiteboard [61] is presented in Section 6.…”