The increased usage of Internet of Things devices at the network edge and the proliferation of microservice-based applications create new orchestration challenges in Edge computing. These include detecting overutilized resources and scaling out overloaded microservices in response to surging requests. This work presents ADApt, an extension of the ADA-PIPE tool developed in the DataCloud project, using the monitoring data related to Edge devices, detecting the utilization-based anomalies of resources (e.g., processing or memory), investigating the scalability in microservices, and adapting the application executions. To reduce the overutilization bottleneck, we first explore monitored devices executing microservices over various time slots, detecting overutilization-based processing events, and scoring them. Thereafter, based on the memory requirements, ADApt predicts the processing requirements of the microservices and estimates the number of replicas running on the overutilized devices. The prediction results show that the gradient boosting regression-based replica prediction reduces the MAE, MAPE, and RMSE compared to other models. Moreover, ADApt can estimate the number of replicas for each microservice close to the actual data without any prediction and reduce the CPU utilization of the device by 14 % − 28 %.

Recent progress in large language models and access to large-scale robotic datasets has sparked a paradigm shift in robotics models transforming them into generalists able to adapt to various tasks, scenes, and robot modalities. A large step for the community are open Vision Language Action models which showcase strong performance in a wide variety of tasks. In this work, we study the visual generalization capabilities of three existing robotic foundation models, and propose a corresponding evaluation framework. Our study shows that the existing models do not exhibit robustness to visual out-of-domain scenarios. This is potentially caused by limited variations in the training data and/or catastrophic forgetting, leading to domain limitations in the vision foundation models. We further explore OpenVLA, which uses two pre-trained vision foundation models and is, therefore, expected to generalize to out-of-domain experiments. However, we showcase catastrophic forgetting by DINO-v2 in OpenVLA through its failure to fulfill the task of depth regression. To overcome the aforementioned issue of visual catastrophic forgetting, we propose a gradual backbone reversal approach founded on model merging. This enables OpenVLA - which requires the adaptation of the visual backbones during initial training - to regain its visual generalization ability. Regaining this capability enables our ReVLA model to improve over OpenVLA by a factor of 77% and 66% for grasping and lifting in visual OOD tasks. Comprehensive evaluations, episode rollouts and model weights are available on the ReVLA Page

Cloud computing has become an increasingly popular choice for businesses and individuals due to its flexibility, scalability, and convenience; however, the rising cost of cloud resources has become a significant concern for many. The pay-per-use model used in cloud computing means that costs can accumulate quickly, and the lack of visibility and control can result in unexpected expenses. The cost structure becomes even more complicated when dealing with hybrid or multi-cloud environments. For businesses, the cost of cloud computing can be a significant portion of their IT budget, and any savings can lead to better financial stability and competitiveness. In this respect, it is essential to manage cloud costs effectively. This requires a deep understanding of current resource utilization, forecasting future needs, and optimising resource utilization to control costs. To address this challenge, new tools and techniques are being developed to provide more visibility and control over cloud computing costs. In this respect, this paper explores a graph-based solution for modelling cost elements and cloud resources and potential ways to solve the resulting constraint problem of cost optimisation. We primarily consider utilization, cost, performance, and availability in this context. Such an approach will eventually help organizations make informed decisions about cloud resource placement and manage the costs of software applications and data workflows deployed in single, hybrid, or multi-cloud environments.

This paper describes the DATACLOUDDSL language and the DEF-PIPE tool for describing Big Data pipelines. DAT-ACLOUDDSL has both a textual and a visual form and supports requirements obtained both from analyzing existing data pipeline specification tools and from interviews with relevant industrial actors. Particularly, DATACLOUDDSL supports (i) separation of concerns between design and run-time issues, (ii) reuse of previously developed pipeline steps and pipelines in designing new pipelines, (iii) flexible data transfer between pipelines steps and containerization of pipelines and pipeline steps, and (iv) integration of description and simulation components in Big Data pipeline orchestration systems. Additionally, it provides an interface to the discovery and deployment tools of the DataCloud toolbox.

Big data pipelines are becoming increasingly vital in a wide range of data intensive application domains such as digital healthcare, telecommunication, and manufacturing for efficiently processing data. Data pipelines in such domains are complex and dynamic and involve a number of data processing steps that are deployed on heterogeneous computing resources under the realm of the Edge-Cloud paradigm. The processes of testing and simulating big data pipelines on heterogeneous resources need to be able to accurately represent this complexity. However, since big data processing is heavily resource-intensive, it makes testing and simulation based on historical execution data impractical. In this paper, we introduce the SIM - PIPE Dry Runner approach - a dry run approach that deploys a big data pipeline step by step in an isolated environment and executes it with sample data; this approach could be used for testing big data pipelines and realising practical simulations using existing simulators.

Organisations possess and continuously generate huge amounts of static and stream data, especially with the proliferation of Internet of Things technologies. Collected but unused data, i.e., Dark Data, mean loss in value creation potential. In this respect, the concept of Computing Continuum extends the traditional more centralised Cloud Computing paradigm with Fog and Edge Computing in order to ensure low latency pre-processing and filtering close to the data sources. However, there are still major challenges to be addressed, in particular related to management of various phases of Big Data processing on the Computing Continuum. In this paper, we set forth an ecosystem for Big Data pipelines in the Computing Continuum and introduce five relevant real-life example use cases in the context of the proposed ecosystem.

The development of the Edge computing paradigm shifts data processing from centralised infrastructures to heterogeneous and geographically distributed infrastructure. Such a paradigm requires data processing solutions that consider data locality in order to reduce the performance penalties from data transfers between remote (in network terms) data centres. However, existing Big Data processing solutions have limited support for handling data locality and are inefficient in processing small and frequent events specific to Edge environments. This paper proposes a novel architecture and a proof-of-concept implementation for software container-centric Big Data workflow orchestration that puts data locality at the forefront. Our solution considers any available data locality information by default, leverages long-lived containers to execute workflow steps, and handles the interaction with different data sources through containers. We compare our system with Argo workflow and show significant performance improvements in terms of speed of execution for processing units of data using our data locality aware Big Data workflow approach.

Hand-crafting generalised decision-making rules for real-world urban autonomous driving is hard. Alternatively, learning behaviour from easy-to-collect human driving demonstrations is appealing. Prior work has studied imitation learning (IL) for autonomous driving with a number of limitations. Examples include only performing lane-following rather than following a user-defined route, only using a single camera view or heavily cropped frames lacking state observability, only lateral (steering) control, but not longitudinal (speed) control and a lack of interaction with traffic. Importantly, the majority of such systems have been primarily evaluated in simulation - a simple domain, which lacks real-world complexities. Motivated by these challenges, we focus on learning representations of semantics, geometry and motion with computer vision for IL from human driving demonstrations. As our main contribution, we present an end-to-end conditional imitation learning approach, combining both lateral and longitudinal control on a real vehicle for following urban routes with simple traffic. We address inherent dataset bias by data balancing, training our final policy on approximately 30 hours of demonstrations gathered over six months. We evaluate our method on an autonomous vehicle by driving 35km of novel routes in European urban streets.

Big Data processing involves handling large and complex data sets, incorporating different tools and frameworks as well as other processes that help organisations make sense of their data collected from various sources. This set of operations, referred to as Big Data workflows, require taking advantage of the elasticity of cloud infrastructures for scalability. In this paper, we present the design and prototype implementation of a Big Data workflow approach based on the use of software container technologies and message-oriented middleware (MOM) to enable highly scalable workflow execution. The approach is demonstrated in a use case together with a set of experiments that demonstrate the practical applicability of the proposed approach for the scalable execution of Big Data workflows. Furthermore, we present a scalability comparison of our proposed approach with that of Argo Workflows - one of the most prominent tools in the area of Big Data workflows.

Binary sequences are widely used in various practical fields, such as telecommunications, radar technology, navigation, cryptography, measurement sciences, biology or industry. In this letter, a method to generate long binary sequences (LBS) with low peak sidelobe level (PSL) value is proposed. Having an LBS with length n, both the time and memory complexities of the proposed algorithm are O(n). During our experiments, we repeatedly reach better PSL values than the currently known state of art constructions, such as Legendre sequences, with or without rotations, Rudin-Shapiro sequences or m-sequences, with or without rotations, by always reaching a record-breaking PSL values strictly less than √n. Furthermore, the efficiency and simplicity of the proposed method are particularly beneficial to the lightweightness of the implementation, which allowed us to reach record-breaking PSL values for less than a second.

Simple and efficient algorithm based on heuristic search by shotgun hill climbing to construct binary sequences with small peak sidelobe levels (PSL) is suggested. The algorithm is applied for generation of binary sequences of lengths between 106 and 300. Improvements are obtained in almost half of the considered lengths while for the rest of the lengths, binary sequences with the same PSL values as reported in the state-of-the-art publications are found.

MOTIVATION: The online resources in the life sciences are characterized by a great fragmentation and one of the pressing issues of bioinformatics is making the integration of these resources a smoother and more flexible process than it is currently. Here we present i-cite, a browser extension, which implements a client-side model of integration which improves the navigation within the rapidly increasing life science literature and links terms from it to corresponding non-textual data. AVAILABILITY: http://i-cite.org.

Currently biomedical literature is largely disconnected from its metadata. While there are freely accessible centralised metadata repositories the publications themselves are split among a large number of repositories. We address this problem by harvesting freely accessible biomedical publications from the Web and integrating them with the corresponding metadata. The system involves title recognition applied on the harvested publications using knowledge-based algorithm and a fuzzy match between the extracted title and the metadata records using edit distance metric. So far we were able to locate +300.000 publications on the Web and achieve +96% precision and nearly 85% recall on a random sample of 250 documents harvested from the Web.

We present a system for extracting citation data from Pubmed-indexed papers available online based on a knowledge-based algorithm. We achieve nearly 92% accuracy on a sample of 156 papers from 78 different journals. We describe the issues faced, our approach, the results achieved and the future directions of our work.