Secure Neural Network Inference (SNNI) protocols, vital for privacy-preserving AI, face substantial computational and communication overhead. Dynamic Early-Exit (EE) networks could help decrease the overhead, but existing SNNI protocols do not support such networks. We introduce QUOKKA, the first system to enable SNNI for confidence-based EE neural networks using secure Multi-Party Computation. QUOKKA addresses the challenges of dynamic decision-making and sensitive intermediary result handling in SNNI. Implemented with EENet and CrypTen, QUOKKA achieves 2–5 × acceleration over traditional SNNI without a decrease in accuracy. Our findings demonstrate practical, highly efficient, and privacy-preserving SNNI for dynamic AI, paving the way for broader Machine-Learning-as-a-Service deployment.

Secure Neural Network Inference (SNNI) protocols enable privacy-preserving inference by ensuring the confidentiality of inputs, model weights, and outputs. However, large neural networks, particularly Transformers, face significant challenges in SNNI due to high computational costs and slow execution, as these networks are typically optimized for accuracy rather than secure inference speed. We present COLIBRI, a novel approach that optimizes neural networks for efficient SNNI using Neural Architecture Search (NAS). Unlike prior methods, COLIBRI directly incorporates SNNI execution time as an optimization objective, leveraging a prediction model to estimate execution time without repeatedly running costly SNNI protocols during NAS. Our results on Cityscapes, a complex image segmentation task, show that COLIBRI reduces SNNI execution time by 26–33

The rapid growth of Artificial Intelligence (AI) applications, particularly through the widespread adoption of Large Language Models (LLMs), has caused an unprecedented growth in computing and network infrastructures. Current infrastructure expansion cannot keep pace, resulting in suboptimal performance. This creates an urgent need for network automation capable of dynamically orchestrating services and exploiting all available resources. Manual optimization processes are slow, error-prone, and unable to meet the requirements of complex, multi-domain, and data-intensive networks. A fundamental challenge is the absence of a universal optimization algorithm that performs effectively across all scenarios. In this paper, we present preliminary work on an LLM-based optimization algorithm selection framework for multi-domain, high-performance networks orchestration. The proposed framework utilizes LLM-generated descriptive embeddings of algorithms, network state logs, and service requests to identify the most suitable optimization method from a pool of algorithms, curating optimization to the current scenario.

The escalating demands of scientific collaborations necessitate advanced networking for deterministic, secure, and orchestrated services across multiple administrative domains. The Software-Defined Network for End-to-end Networked Science at the Exascale (SENSE) paradigm addresses these needs through intent-based networking and multi-domain orchestration. This paper evaluates SENSE's performance on a comprehensive multi-domain testbed, including GNA-G AutoGOLE, the National Research Platform (NRP), FABRIC, and production LHC CMS infrastructure. Our results demonstrate that intent-based service requests are successfully translated into network configurations, with average provisioning times of 183 seconds for simple services and 290 seconds for complex multi-domain workflows. Performance monitoring confirms that SENSE maintains guaranteed bandwidth allocations, enabling higher-priority data flows to complete significantly faster than in best-effort scenarios. This capability transforms the network into a first-class schedulable resource, optimizing scientific workflows by prioritizing data criticality and moving beyond best-effort limitations to achieve predictable and efficient data movement for data-intensive scientific endeavors.

According to the European Union Aviation Safety Agency (EASA), AI-based algorithms, combined with extensive fleet data, could enable early detection of potential engine failures, leading to proactive predictive maintenance in air travel. At a global level, the Independent Data Consortium for Aviation (IDCA) recognizes the potential of collaborative data sharing in the airline industry. However, data ownership-related issues, such as privacy, intellectual property, and regulatory compliance, pose significant obstacles to realizing the vision of combining fleet data to improve predictive maintenance algorithms. In this paper, we use NASA’s Turbofan Jet Engine Dataset (N-CMAPSS) to demonstrate how airlines could leverage the power of Federated Learning (FL) and microservices, to collaboratively train a global Machine-Learning (ML) model that can enable airline companies to utilize their data for predictive maintenance, while maintaining control.

External supervision services play an important role in combating corruption by detecting potential collusive bribery. This work aims at studying the dynamics of collusive bribery when participants have the option of engaging the external supervision services. To do so, we construct a basic model where collusive bribery can happen between the defecting participants who aim to escape from a punishment by offering a bribe to rule enforcers who monitor interactions among all participants. Among rule enforcers, only the corrupt ones accept the bribe and ignore the violations. The cooperative participants can engage the external supervision service at a certain cost to avoid the risk of potential collusive bribery. Under the framework of evolutionary game theory, we find that a higher initial fraction of honest enforcers is more likely to lead to a trusting cooperating equilibrium. We also find that, when allowing random exploration of available strategies, increasing the exploration rate of rule enforcers is effective in combating corruption for both infinite and finite populations. Lastly, we find that minimizing the cost of external supervision services is not always good. When the system evolves into a cooperating equilibrium, a low cost of external supervision service induces unnecessary costs of seeking external supervision. When the strategy profiles exhibit stable oscillations, there exists an optimal cost of external supervision, considering the trade-off between minimizing the chance of exposing cooperative participants to collusive bribery and strengthening the punishment on the corrupt enforcers. Premised on the results, we discuss practical management suggestions.& COPY; 2023 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ )

In recent years, blockchain has gained widespread attention as an emerging technology for decentralization, transparency, and immutability in advancing online activities over public networks. As an essential market process, auctions have been well studied and applied in many business fields due to their efficiency and contributions to fair trade. Complementary features between blockchain and auction models trigger a great potential for research and innovation. On the one hand, the decentralized nature of blockchain can provide a trustworthy, secure, and cost-effective mechanism to manage the auction process; on the other hand, auction models can be utilized to design incentive and consensus protocols in blockchain architectures. These opportunities have attracted enormous research and innovation activities in both academia and industry; however, there is a lack of an in-depth review of existing solutions and achievements. In this paper, we conduct a comprehensive state-of-the-art survey of these two research topics. We review the existing solutions for integrating blockchain and auction models, with some application-oriented taxonomies generated. Additionally, we highlight some open research challenges and future directions towards integrated blockchain-auction models.

Context. As radio telescopes increase in sensitivity and flexibility, so do their complexity and data rates. For this reason, automated system health management approaches are becoming increasingly critical to ensure nominal telescope operations. Aims. We propose a new machine-learning anomaly detection framework for classifying both commonly occurring anomalies in radio telescopes as well as detecting unknown rare anomalies that the system has potentially not yet seen. To evaluate our method, we present a dataset consisting of 6708 autocorrelation-based spectrograms from the Low Frequency Array (LOFAR) telescope and assign ten different labels relating to the system-wide anomalies from the perspective of telescope operators. This includes electronic failures, miscalibration, solar storms, network and compute hardware errors, among many more. Methods. We demonstrate how a novel self-supervised learning (SSL) paradigm, that utilises both context prediction and reconstruction losses, is effective in learning normal behaviour of the LOFAR telescope. We present the Radio Observatory Anomaly Detector (ROAD), a framework that combines both SSL-based anomaly detection and a supervised classification, thereby enabling both classification of both commonly occurring anomalies and detection of unseen anomalies. Results. We demonstrate that our system works in real time in the context of the LOFAR data processing pipeline, requiring <1ms to process a single spectrogram. Furthermore, ROAD obtains an anomaly detection F-2 score of 0.92 while maintaining a false positive rate of 2%, as well as a mean per-class classification F-2 score of 0.89, outperforming other related works.

The increasing value of data and the emergence of programmable infrastructures have paved the way for col-laborative multi-domain applications across industries such as healthcare and airlines. However, such collaborations come with significant challenges, including application coordination, incentivization, and validation of execution. In this paper, we propose a novel solution that leverages blockchain technology and utilizes Petri nets for workflow modeling. Our approach involves implementing a smart contract-based workflow coordinator on the blockchain and employing a three-layered architecture to coordinate off-chain tasks. Additionally, we demonstrate the use of Petri nets for modeling economy tokens, which serve as incentives to foster collaboration among workflow parties. To validate our solution, we present a proof of concept through a simulated use case involving a multi-domain workflow for mitigating a DDoS attack. In this use case, domains collaborate by blocking offending IPs, incentivized by acquired tokens required to invoke workflows.

AbstractModern science increasingly works with large amount of data, which are heterogeneous, are distributed, and require special infrastructure for data collection, storage, processing, and visualization. Science digitalization, likewise industry digitalization, is facilitated by the explosive development of digital technologies and cloud-based infrastructure technologies and services. This paper attempts to understand impact and new requirements to the future Scientific Data Infrastructure imposed by growing science digitalization. The paper presents two lines of analysis: one is a retrospective analysis related to the European Research Infrastructure (RI) development stages and timeline from centralized to distributed and current Federated Interoperable; another line provided analysis of digital technology trends and identified what technologies will impact the future Scientific Data Infrastructure (SDI). Based on this analysis, the paper proposes a vision for the future RI Platform as a Service (PRIaaS) that incorporates recent digital technologies and enables platform and ecosystem model for future science. Notably the proposed PRIaaS adopts TMForum Digital Platform Reference Architecture (DPRA) that will simplify building and federating domain-specific RIs while focusing on the domain-specific data value chain with data protection and policy-based management by design.

Incentives are usually introduced by the regulator entity (third-party), to promote cooperation in a market. The implementation of incentives is always costly and thus might fail to be enforced sustainably. This work aims at exploring the effects of incentives from an institutional perspective, while coping with the scenario where the third-party is part of the system but not composed by players. The evolutionary game theory (EGT) framework is applied to identify the incentives that lead to pure cooperation. In contrast to traditional EGT, this paper introduces an elimination mechanism that can reduce the market size. The incentives identified in the EGT analysis are further examined in simulation experiments which measure the market size, affluence and sustainability. The findings show: (1) light punishment leads to a reduction of the market size, yet heavier punishment is beneficial to the market size and wealth; (2) mixed incentives will generally lead to different wealth of the third party and of the participants. While under moderate strength, the wealth of both parties is the same and their overall wealth is maximal; (3) for sustainability, pure punishment (resp. reward) is sustainable (resp. unsustainable), the sustainability of mixed incentives depends on both their strength and agents’ rationality level.

In this paper, we explore the data recovery procedures from ${e} \cdot $ MMCs. The ${e} \cdot $ MMC is one of the “managed” flash memory devices that are popularly used in modern digital devices as their storage media. The ${e} \cdot $ MMC, which consists of flash memory and the flash memory controller, optimizes the data input/output between the host device and the non-volatile memory through its standardized protocol. Its standardized structure and protocol makes forensic physical data acquisition simpler than handling the raw flash memory. However, its secure data purging features, such as Secure Erase and Sanitize, make data recovery from ${e} \cdot $ MMC a challenging task. In this research, we investigate inside the ${e} \cdot $ MMCs, and evaluate advanced data recovery procedures. By reverse engineering the structures of ${e} \cdot $ MMCs and accessing the internal flash memory, we discover that securely erased data is still recoverable from the internal flash memory. In some models, more than 99% of the securely erased data can still be recoverable by accessing the flash memory inside the ${e} \cdot $ MMCs. The data extraction method, along with experimental data recovery evaluation, will be explored in this paper.

Industrial applications often require federated cloud services from multiple providers to improve reliability and flexibility. Traditional selection methods through auctions usually involve a centralized auctioneer to coordinate the auction procedure. Blockchain and smart contracts provide a decentralized mechanism to automate the cloud auction process; however, existing solutions fail in the selection of the most suitable providers and the violation detection of the signed auction agreements, which are also known as service-level agreements (SLAs). To tackle these problems, we propose an integrated auction model using Bayesian game theory and blockchain techniques. The proposed model is enhanced with two Bayesian Nash Equilibriums (BNEs); the first BNE enables the selection of cost-effective providers to construct the federated cloud services, while the second BNE ensures consistent and trustworthy monitoring of federated SLAs. Moreover, a timed message submission (TMS) algorithm is proposed to protect the auction privacy during the message submission phase. This paper validates the equilibrium results of two BNEs and implements the proposed model on the Ethereum blockchain. The analytical and experimental results demonstrate the feasibility, trustworthiness, and cost-effectiveness of our model.

In recent years, blockchain has gained widespread attention as an emerging technology for decentralization, transparency, and immutability in advancing online activities over public networks. As an essential market process, auctions have been well studied and applied in many business fields due to their efficiency and contributions to fair trade. Complementary features between blockchain and auction models trigger a great potential for research and innovation. On the one hand, the decentralized nature of blockchain can provide a trustworthy, secure, and cost-effective mechanism to manage the auction process; on the other hand, auction models can be utilized to design incentive and consensus protocols in blockchain architectures. These opportunities have attracted enormous research and innovation activities in both academia and industry; however, there is a lack of an in-depth review of existing solutions and achievements. In this paper, we conduct a comprehensive state-of-the-art survey of these two research topics. We review the existing solutions for integrating blockchain and auction models, with some application-oriented taxonomies generated. Additionally, we highlight some open research challenges and future directions towards integrated blockchain-auction models.

A Digital Data Marketplace (DDM) is a digital infrastructure to facilitate policy-governed data sharing in a secure and trustworthy manner with container-based virtualization technologies. An intrusion detection systems (IDS) is essential to enforce the policies. We propose a real-time intrusion detection system that monitors and analyzes the Linux-kernel system calls of a running container. We adopt the One-Class Support Vector Machine (OC-SVM) to detect anomalies. The training data of the OC-SVM algorithm is collected and sanitized in a secure environment. We evaluate the detection capability of our proposed system against modern attacks, e.g. Machine Learning (ML) adversarial attacks, with a customized attack dataset. In addition, we investigate the influence of various feature extraction methods, kernel functions and segmentation length with four metrics. Our experimental results show that we can achieve a low FPR, with a worst case of 0.12, and a TPR of 1 for most attacks, when we adopt the term-frequency feature extraction method and we choose segmentation length of 30000. Furthermore, the optimal kernel functions depend on the concrete application being examined.

This research aims to improve availability in Industrial Control Systems (ICS) us-ing Software-Defined Networking (SDN). Programmable Logic Controllers (PLC) are the brains of ICSs. If these critical devices would become unavailable, it could lead to operational disruptions and widespread damage. Network Function Virtualization (NFV) could be used to create a more agile and lower cost infrastructure, since hardware could be virtualized. On top of that, Software-Defined Networking (SDN) can be used to manage networks by separating the control plane from the data plane. Our research has shown that SDN combined with NFV provides efficiency, flexibility and reduces human error since the action taken by the software will be dynamic. This raises the question whether SDN combined with NFV could enhance the availability of ICS environments and thus reduces the risk of disruptions and widespread damage. According to previous research, on average there is a downtime of one hour, since it will take 30 minutes to summon a technician and another 30 minutes to repair the faulty network module (e.g. switch). This research has tried to answer this question by implementing three SDN driven scenarios: re-routing of the traffic, redeployment of network hardware and recreation of specific interfaces of the network hardware. Our experiments showed that all three scenarios could be used to improve the availability of ICS environments. However, differences can be observed between these scenarios. Re-routing traffic in case of a switch failure showed to perform the worst which had an average downtime of 5576ms, while the scenario in which a switch would be redeployed showed to perform the best which had an average downtime of only 48ms. However, we were unable to obtain 100% availability. Since the NFV function was written in Python, the efficiency of the code did not allow us to set an interval small enough to detect failures fast enough. Moreover, the performance of the machine the code was running on, could also have played a role. Using SDN and NFV will result in complexity in an ICS environment. One would trade simplicity for the speed of recovery in case of a hardware failure, availability and also the ability of having more machines running while using less hardware.

Wi-Fi Protected Access 3 (WPA3) became a mandatory part of the Wi-Fi certification on July 1st 2020. Therefore, the adoption rate of WPA3 is expected to grow soon. In this paper, we focus on WPA3 personal transition mode, in particular the security of this mode. We argue that transition mode is a requirement in home environments for the foreseeable future. We investigate whether it is possible to secure a WPA3 personal transition mode network in such a way that downgrade attacks are not feasible. We find that even with the security recommendations that the Wi-Fi Alliance recently issued for WPA3, common implementations running in transition mode can still be downgraded to WPA2. In our experiments, we can see that there are differences between WPA3 implementations in terms of security. The Wi-Fi Alliance has already announced upcoming additions to the WPA3 standard. These additions offer essential improvements to the security of WPA3 personal transition mode networks. We believe that the WPA3 certification should be extended to include the recently announced additions to WPA3. In addition to this, we make several recommendations to ensure the safe operation of WPA3. Together these changes will resolve most of the implementation differences we observed. Furthermore, we argue that mutual authentication is an essential stepping stone towards a more secure Wi-Fi ecosystem and discuss two mechanisms.

A Digital Data Marketplace (DDM) facilitates secure and trustworthy data sharing among multiple parties. For instance, training a machine learning (ML) model using data from multiple parties normally contributes to higher prediction accuracy. It is crucial to enforce the data usage policies during the execution stage. In this paper, we propose a methodology to distinguish programs running inside containers by monitoring system calls sequence externally. To support container portability and the necessity of retraining ML models, we also investigate the stability of the proposed methodology in 7 typical containerized ML applications over different execution platform OSs and training data sets. The results show our proposed methodology can distinguish between applications over various configurations with an average classification accuracy of 93.85%, therefore it can be integrated as an enforcement component in DDM infrastructures.