Adaptive Model Scheduling for Resource-efficient Data Labeling

Labeling data (e.g., labeling the people, objects, actions, and scene in images) comprehensively and efficiently is a widely needed but challenging task. Numerous models were proposed to label various data and many approaches were designed to enhance the ability of deep learning models or accelerate them. Unfortunately, a single machine-learning model is not powerful enough to extract various semantic information from data. Given certain applications, such as image retrieval platforms and photo album management apps, it is often required to execute a collection of models to obtain sufficient labels. With limited computing resources and stringent delay, given a data stream and a collection of applicable resource-hungry deep-learning models, we design a novel approach to adaptively schedule a subset of these models to execute on each data item, aiming to maximize the value of the model output (e.g., the number of high-confidence labels). Achieving this lofty goal is nontrivial since a model’s output on any data item is content-dependent and unknown until we execute it. To tackle this, we propose an Adaptive Model Scheduling framework, consisting of (1) a deep reinforcement learning-based approach to predict the value of unexecuted models by mining semantic relationship among diverse models, and (2) two heuristic algorithms to adaptively schedule the model execution order under a deadline or deadline-memory constraints, respectively. The proposed framework does not require any prior knowledge of the data, which works as a powerful complement to existing model optimization technologies. We conduct extensive evaluations on five diverse image datasets and 30 popular image labeling models to demonstrate the effectiveness of our design: our design could save around 53% execution time without loss of any valuable labels.

Delay Aware Accident Detection and Response System Using IoT

Vehicle to vehicle communication and vehicle to roadside sensor communication are introduced with the help of hybrid ITS safety architecture.It require major investments for installation, purchase and maintenance for implementation. It represents a costbeneficial solution and can leverage the deployment of the system as a whole for roadside wireless sensor and networking technology . This paper proposepostaccident inquiry and accident precaution. A framework and convention engineering with a completely disseminated idea for secure capacity of sensor information and proficient is incorporated. This is the blend of information stockpiling and committed side of the road units as an incorporated organization component for correspondence for arrangement. At last, it depicts the product stages for vehicle on-board units and sensor hubs. model execution and exploratory proving ground including equipment.

An executable axiomatization of the REA2 ontology

The formalization of the REA2 ontology presented in this paper offers a minimal set of operationalized semantics for a single white-box model relevant to all business stakeholders independent of their role or involvement in economic activities. This paper's theoretical innovations are the use of MERODE to model increment and decrement semantics as fundamental stand-alone concepts that simultaneously affect economic resources, event, agents and the semantics of the stock-flow, participation and ownership associations and the formalization of the REA axioms as executable finite state machines. MERODE's possibilities for model execution through fast prototyping allowed validation through the modeling of an archetypical exchange scenario. Both innovations contribute to the reliability of a generic semantic model for finance and logistics in both the traditional as well as the sharing economy, thus promoting traceability and accountability in value networks and supply chains supported by both centralized and decentralized ledger technologies.

The role and value of distributed precipitation data in hydrological models

This study investigates the role and value of distributed rainfall for the runoff generation of a mesoscale catchment (20 km2). We compare four hydrological model setups and show that a distributed model setup driven by distributed rainfall only improves the model performances during certain periods. These periods are dominated by convective summer storms that are typically characterized by higher spatiotemporal variabilities compared to stratiform precipitation events that dominate rainfall generation in winter. Motivated by these findings, we develop a spatially adaptive model that is capable of dynamically adjusting its spatial structure during model execution. This spatially adaptive model allows the varying relevance of distributed rainfall to be represented within a hydrological model without losing predictive performance compared to a fully distributed model. Our results highlight that spatially adaptive modeling has the potential to reduce computational times as well as improve our understanding of the varying role and value of distributed precipitation data for hydrological models.

Modeling and Analyzing Context-Sensitive Changes during Runtime

For collaborative embedded systems, it is essential to consider not only the behavior of each system and the interaction between systems, but also the interaction of systems with their often dynamic and unknown context.In this chapter, we present a solution approach based on process building blocks— describing both the modelling approach as well as the model execution approach—for engineering and operation to achieve the goal of developing systems that deal with dynamics in their open context at runtime by re-using the models from the engineering phase.

The EurValve model execution environment

The goal of this paper is to present a dedicated high-performance computing (HPC) infrastructure which is used in the development of a so-called reduced-order model (ROM) for simulating the outcomes of interventional procedures which are contemplated in the treatment of valvular heart conditions. Following a brief introduction to the problem, the paper presents the design of a model execution environment, in which representative cases can be simulated and the parameters of the ROM fine-tuned to enable subsequent deployment of a decision support system without further need for HPC. The presentation of the system is followed by information concerning its use in processing specific patient cases in the context of the EurValve international collaboration.

Hybrid deep neural network for Bangla automated image descriptor

Automated image to text generation is a computationally challenging computer vision task which requires sufficient comprehension of both syntactic and semantic meaning of an image to generate a meaningful description. Until recent times, it has been studied to a limited scope due to the lack of visual-descriptor dataset and functional models to capture intrinsic complexities involving features of an image. In this study, a novel dataset was constructed by generating Bangla textual descriptor from visual input, called Bangla Natural Language Image to Text (BNLIT), incorporating 100 classes with annotation. A deep neural network-based image captioning model was proposed to generate image description. The model employs Convolutional Neural Network (CNN) to classify the whole dataset, while Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) capture the sequential semantic representation of text-based sentences and generate pertinent description based on the modular complexities of an image. When tested on the new dataset, the model accomplishes significant enhancement of centrality execution for image semantic recovery assignment. For the experiment of that task, we implemented a hybrid image captioning model, which achieved a remarkable result for a new self-made dataset, and that task was new for the Bangladesh perspective. In brief, the model provided benchmark precision in the characteristic Bangla syntax reconstruction and comprehensive numerical analysis of the model execution results on the dataset.

Supporting Multi-cloud Model Execution with VLab

<p>The VLab (Virtual Laboratory), developed in the context of the European projects ECOPOTENTIAL and ERA-PLANET, is a cloud-based platform to support the activity of environmental scientists in sharing their models. The main challenges addressed by VLab are: (i) minimization of interoperability requirements in the process of model porting (i.e. to simplify as much as possible the process of publishing and sharing a model for model developers) and (ii) support multiple programming languages and environments (it must be possible porting models developed in different programming languages and which use an arbitrary set of libraries).</p><p>In this presentation we describe how VLab supports a multi-cloud deployment approach and the benefits.</p><p>In this presentation we describe VLab architecture and, in particular, how this enables supporting a multi-cloud deployment approach.</p><p>Deploying VLab on different cloud environments allows model execution where it is most convenient, e.g. depending on the availability of required data (move code to data).</p><p>This was implemented in the web application for Protected Areas, developed by the Joint Research Centre of the European Commission (EC JRC) in the context of the EuroGEOSS Sprint to Ministerial activity and demonstrated at the last GEO-XVI Plenary meeting in Canberra. The web application demonstrates the use of Copernicus Sentinel data to calculate Land Cover and Land Cover change in a set of Protected Areas belonging to different ecosystems. Based on user’s selection of satellite products to use, the different available cloud platforms where to run the model are presented along with their data availability for the selected products. After the platform selection, the web application utilizes the VLab APIs to launch the EODESM (Earth Observation Data for Ecosystem Monitoring) model (Lucas and Mitchell, 2017), monitoring the execution status and retrieve the output.</p><p>Currently, VLab was experimented with the following cloud platforms: Amazon Web Services, three of the 4+1 the Coperncius DIAS platforms (namely: ONDA, Creodias and Sobloo) and the European Open Science Cloud (EOSC).</p><p>Another possible scenario empowered by this multi-platform deployment feature is the possibility to let the user choose the computational platform and utilize her/his credentials to request the needed computational resources. Finally, it is also possible to exploit this feature for benchmarking different cloud platforms with respect to their performances.</p><p> </p><p>References</p><p>Lucas, R. and A. Mitchell (2017). "Integrated Land Cover and Change Classifications"The Roles of Remote Sensing in Nature Conservation, pp. 295–308.</p><p> </p>

Discrete Event Simulation of Clients Flow in Ante-natal Clinic

Clients (expecting mothers) wait for hours in ante-natal clinic to receive medical service – waiting before, during or after being served. This study deals with a dynamic queuing system. Results of the study evaluate the effectiveness of queuing simulation model by identifying the ante-natal clinic queuing system parameters in terms of server utilization, usage, and clients flow time. The study uses the Simmer package in R for Discrete-event Simulation of the clients' flow in the system. The study showed that the resources are highly utilized with a bottleneck at the Doctors station, with constant service time for all clients, and long waiting time in the system. By replicating the parameters or replicate the model execution, once, with different initial conditions (by adding resources) and then performing another simulation over the output, the result showed that the resources are utilized with no bottlenecks at each server station, constant activity and flow time for all clients (expecting mothers). Hence, the model has proved to be accurate and efficient. This will help the clinic to utilize the resources and reduce long flow time.