Abstract The more critical infrastructures (CIs) being digitized, the more vulnerable they are regarding cyber security attacks. Digitisation-leveraging technologies in the Internet of Things (IoT) and Cyber-Physical Systems (CPS) have been largely adopted for CIs, along with the Digital Twin (DT) paradigm. However, the distributed and heterogeneous nature of IoT or CPS poses significant challenges in safeguarding against diverse attack surfaces, including physical devices, network infrastructures, and third-party integration. To tackle these challenges, we propose an AI-driven DT-based security orchestration automation and response framework (SOAR4BC). Gathering system contexts from the DT in combination with security intelligence from the security tools gives us a holistic context for SOAR, which has not been seen in the existing approaches. We leverage this holistic context into the decision-making core, which utilizes advanced algorithms, like deep reinforcement learning, to generate adaptation recommendations based on incident alerts, risk assessments, and system state observations. By rigorously evaluating tampered data and distributed denial of service (DDoS) scenarios, we validate the SOAR4BC framework’s efficacy in handling security incidents leveraging digital twin environments. We further demonstrate real-world applicability through false-data injection and DoS attacks on an operational electric-vehicle charging testbed, confirming the practical effectiveness of SOAR4BC in securing critical infrastructures. Together, these results establish SOAR4BC as a robust and explainable AI-driven SOAR framework that advances the use of digital twins for cybersecurity in IoT and CPS ecosystems, offering actionable contributions for both research and industrial deployment.

The rapid advancement of Large Language Models (LLMs) has opened new possibilities for intelligent multi-agent systems capable of autonomously performing complex tasks. To build such systems, LLMs can be leveraged for task-solving, tool interaction, and code generation but at the same time their costs and unpredictability have to be properly managed. To do so this paper introduces COPMA, a model-based approach to enabling continuous human-LLM co-programming of multi-agent LLM applications. COPMA uses feature-block models to track application features and their implementations as agents and code blocks. Supported by co-programming patterns, de-velopers are guided in constructing, refining, and refactoring feature implementations via trial-and-errors with LLM agents, leveraging their feedback, suggestions, and code examples. The patterns guide the shift of feature implementations between agents and code to balance flexibility, predictability, and cost. Our experience in developing LLM agents for collecting and reviewing medical research papers demonstrates that human-LLM co-programming can reduce development effort to enable rapid prototyping of multi-agent LLM applications.

The reward function is an essential component in robot learning. Reward directly affects the sample and computational complexity of learning, and the quality of a solution. The design of informative rewards requires domain knowledge, which is not always available. We use the properties of the dynamics to produce system-appropriate reward without adding external assumptions. Specifically, we explore an approach to utilize the Lyapunov exponents of the system dynamics to generate a system-immanent reward. We demonstrate that the Sum of the Positive Lyapunov Exponents (SuPLE) is a strong candidate for the design of such a reward. We develop a computational framework for the derivation of this reward, and demonstrate its effectiveness on classical benchmarks for sample-based stabilization of various dynamical systems. It eliminates the need to start the training trajectories at arbitrary states, also known as auxiliary exploration. While the latter is a common practice in simulated robot learning, it is unpractical to consider to use it in real robotic systems, since they typically start from natural rest states such as a pendulum at the bottom, a robot on the ground, etc. and can not be easily initialized at arbitrary states. Comparing the performance of SuPLE to commonly-used reward functions, we observe that the latter fail to find a solution without auxiliary exploration, even for the task of swinging up the double pendulum and keeping it stable at the upright position, a prototypical scenario for multi-linked robots. SuPLE-induced rewards for robot learning offer a novel route for effective robot learning in typical as opposed to highly specialized or fine-tuned scenarios. Our code is publicly available for reproducibility and further research.

Building an autonomous intelligent agent capable of carrying out web automation tasks from descriptions in natural language offers a wide range of applications, including software testing, virtual assistants, and task automation in general. However, recent studies addressing this problem often require manually constructing of prior human demonstrations. In this paper, we approach the problem by leveraging the idea of reinforcement learning (RL) with the two-phase mechanism to form an agent using LLMs for automating computer tasks without relying on human demonstrations. We evaluate our LLM-based agent using the MiniWob++ dataset of web-based application tasks, showing that our approach achieves 85% success rate without prior demonstrations. The results also demonstrate the agent's capability of self-improvement through training.

The Internet of Things (IoT) is more and more omnipresent. The greater values of the IoT can be realized by enabling data sharing between different stakeholders. However, one of the biggest challenges is ensuring security and enabling trust for IoT data sharing. In this paper, we identify state-of-the-art (SotA) approaches and techniques for secure IoT data sharing. We present high-level results emphasizing the SotA trend and revealing the most addressed domains, as well as more in-depth details such as procedures and methods used to preserve security in the data sharing environment. The blockchain technology, smart contracts, and InterPlanetary File System (IPFS) are among the most widely used approaches. As today’s solutions explore a more decentralized approach to data sharing, there are several aspects to consider. Based on the findings, we have identified potential research directions for future work, including the differences between public and private blockchains, the combinat (More)

In 2016, a team of earth scientists directly engaged a team of computer scientists to identify cyberinfrastructure (CI) approaches that would speed up an earth science workflow. This paper describes the evolution of that workflow as the two teams bridged CI and an image segmentation algorithm to do large scale earth science research. The Pacific Research Platform (PRP) and The Cognitive Hardware and Software Ecosystem Community Infrastructure (CHASE-CI) resources were used to significantly decreased the earth science workflow's wall-clock time from 19.5 days to 53 minutes. The improvement in wall-clock time comes from the use of network appliances, improved image segmentation, deployment of a containerized workflow, and the increase in CI experience and training for the earth scientists. This paper presents a description of the evolving innovations used to improve the workflow, bottlenecks identified within each workflow version, and improvements made within each version of the workflow, over a three-year time period.

We present a cyber-physical-human distributed computing framework, AquaSCALE, for gathering, analyzing and localizing anomalous operations of increasingly failure-prone community water services. Today, detection of pipe breaks/leaks in water networks takes hours to days. AquaSCALE leverages dynamic data from multiple information sources including IoT (Internet of Things) sensing data, geophysical data, human input, and simulation/modeling engines to create a sensor-simulation-data integration platform that can accurately and quickly identify vul-nerable spots. We propose a two-phase workflow that begins with robust simulation methods using a commercial grade hydraulic simulator - EPANET, enhanced with the support for IoT sensor and pipe failure modelings. It generates a profile of anomalous events using diverse plug-and-play machine learning techniques. The profile then incorporates with external observations (NOAA weather reports and twitter feeds) to rapidly and reliably isolate broken water pipes. We evaluate the two-phase mechanism in canonical and real-world water networks under different failure scenarios. Our results indicate that the proposed approach with offline learning and online inference can locate multiple simultaneous pipe failures at fine level of granularity (individual pipeline level) with high level of accuracy with detection time reduced by orders of magnitude (from hours/days to minutes).

We present a cyber-physical-human (CPHS) distributed computing framework, AquaSCALE, for gathering, analyzing and localizing anomalous operations of increasingly failure-prone community water services. Today, detection of water pipe leaks takes hours to days. AquaSCALE leverages dynamic data from multiple information sources including IoT (Internet of Things) sensing data, geophysical data, human input and simulation/modeling engines to create a sensor-simulation-data integration platform that can locate multiple simultaneous pipe failures at fine level of granularity with high level of accuracy and detection time reduced by orders of magnitude (from hours/days to minutes).