Periodic Reporting for period 1 - CoGNETs (Continuums Of Game NETs: swarm intelligence as information processing (CoGNETs))
Okres sprawozdawczy: 2024-06-01 do 2025-11-30
CoGNETs aims to develop a Middleware Framework that empowers IoT, Edge, and Cloud devices to autonomously organize dynamic IoT-to-Cloud swarm continuums for optimal data processing and seamless service provisioning. The framework enables continuums without fixed or predefined orientations—as commonly assumed today—and instead introduces an on-demand, opportunistic paradigm targeting dynamic swarm continuums. These continuums incorporate both permanent and transient infrastructural elements, such as IoT and Edge devices that can join or leave the swarm on-the-fly.
The core idea for optimizing the Computing Continuum is founded on Asymmetric Multi-Player Competitive Games. These games naturally model collaborative and competitive interactions and can be viewed as extensions of classical swarm intelligence approaches, enriched with self-organization mechanisms such as pricing, bidding, and auctioning. Through these mechanisms, devices (i.e. players) behave rationally, autonomously deciding their actions based on local objectives and global system conditions. In this way, cognitive resources are organized as locally as possible and fully exploited to manage services and data both optimally and securely within self-managed heterogeneous environments. Such environments may include legacy enterprise networks, resource-constrained IoT devices, and Edge-to-Cloud infrastructures characterized by diverse connectivity capabilities and data volumes.
The CoGNETs research concept fuses game-theoretic logic directly into individual devices, transforming each into a cognitive resource capable of self-assessment (pricing) of its onboard data and computational assets. Based on this assessment, devices autonomously decide whether—and how—to participate (bidding) in shared computing spaces (auctioning) in response to user service demands. As a result, swarm continuums are dynamically formed through decentralized online optimization processes executed across participating devices. These processes converge toward mathematically well-defined Computing Equilibria, or golden states, in which service performance is maximized while data and computational resources are utilized efficiently, with no waste. A key strength of the proposed game models is their ability to incorporate additional dimensions of heterogeneity beyond data and computational resources. These include cyber-risk exposure, energy consumption, and system faults, providing additional degrees of freedom to enhance device cognition. This enables CoGNETs not only to optimize service execution but also to holistically safeguard the continuum’s cyber-resilience, digital privacy, and energy efficiency.
The main objectives of CoGNETs are shown below:
Objective #1: To build Intelligent Game Agents of enabling self-organization and decision-making abilities at the Edge and device level, using feature-based heterogeneity models and asymmetric competitive games of optimal data & resource sharing.
Objective #2: To build a Distributed Middleware Framework for coordinating dynamic IoT-to-Cloud swarms of autonomous data processing, using analytical model distribution functions and delocalized federated multi-context Broker architectures.
Objective #3: To build End-to-end Security, Identity, Privacy, and Resilience Mechanisms for addressing swarm-centric threats on all System, Application & AI levels, using low-overhead SSI/DID secured by RISC-V and adversarial shielding.
Objective #4: To build Collaborative Federated Learning Mechanisms of improving the AI service response locally, while exploiting the Edge-Cloud to assist in training accuracy, accelerated via RISC-V and pruning/splitting technologies.
Objective #5: To develop a TRL5 Testbed Deployment that will integrate the outcomes of Obj. #1–#4, and validate its proof of concept over emerging Manufacturing (Industry 4.0) Mobility, Health (Health 4.0) sectors and across their supply-chain.
Objective #6: To identify the EU Social, Ethical, Legal, and Privacy policy aspects of the CoGNETs system, and ensure its promotion to IPCEI and DEP, and to national and international Academic and Industrial communities
The core idea for optimizing the Computing Continuum is founded on Asymmetric Multi-Player Competitive Games. These games naturally model collaborative and competitive interactions and can be viewed as extensions of classical swarm intelligence approaches, enriched with self-organization mechanisms such as pricing, bidding, and auctioning. Through these mechanisms, devices (i.e. players) behave rationally, autonomously deciding their actions based on local objectives and global system conditions. In this way, cognitive resources are organized as locally as possible and fully exploited to manage services and data both optimally and securely within self-managed heterogeneous environments. Such environments may include legacy enterprise networks, resource-constrained IoT devices, and Edge-to-Cloud infrastructures characterized by diverse connectivity capabilities and data volumes.
The CoGNETs research concept fuses game-theoretic logic directly into individual devices, transforming each into a cognitive resource capable of self-assessment (pricing) of its onboard data and computational assets. Based on this assessment, devices autonomously decide whether—and how—to participate (bidding) in shared computing spaces (auctioning) in response to user service demands. As a result, swarm continuums are dynamically formed through decentralized online optimization processes executed across participating devices. These processes converge toward mathematically well-defined Computing Equilibria, or golden states, in which service performance is maximized while data and computational resources are utilized efficiently, with no waste. A key strength of the proposed game models is their ability to incorporate additional dimensions of heterogeneity beyond data and computational resources. These include cyber-risk exposure, energy consumption, and system faults, providing additional degrees of freedom to enhance device cognition. This enables CoGNETs not only to optimize service execution but also to holistically safeguard the continuum’s cyber-resilience, digital privacy, and energy efficiency.
The main objectives of CoGNETs are shown below:
Objective #1: To build Intelligent Game Agents of enabling self-organization and decision-making abilities at the Edge and device level, using feature-based heterogeneity models and asymmetric competitive games of optimal data & resource sharing.
Objective #2: To build a Distributed Middleware Framework for coordinating dynamic IoT-to-Cloud swarms of autonomous data processing, using analytical model distribution functions and delocalized federated multi-context Broker architectures.
Objective #3: To build End-to-end Security, Identity, Privacy, and Resilience Mechanisms for addressing swarm-centric threats on all System, Application & AI levels, using low-overhead SSI/DID secured by RISC-V and adversarial shielding.
Objective #4: To build Collaborative Federated Learning Mechanisms of improving the AI service response locally, while exploiting the Edge-Cloud to assist in training accuracy, accelerated via RISC-V and pruning/splitting technologies.
Objective #5: To develop a TRL5 Testbed Deployment that will integrate the outcomes of Obj. #1–#4, and validate its proof of concept over emerging Manufacturing (Industry 4.0) Mobility, Health (Health 4.0) sectors and across their supply-chain.
Objective #6: To identify the EU Social, Ethical, Legal, and Privacy policy aspects of the CoGNETs system, and ensure its promotion to IPCEI and DEP, and to national and international Academic and Industrial communities
During the first reporting period, CoGNETs established the scientific and technical foundations for cognitive, game-driven computing continuums. Activities combined desk research, system design, theoretical modeling, and initial software development, together with the preparation of the pilot use cases.
Core efforts focused on defining the cognitive computing continuum and translating it into concrete system requirements for the CoGNETs middleware and pilots. Based on expert input and a comprehensive state-of-the-art analysis, the overall CoGNETs architecture was defined, integrating game-theoretic intelligence, collaborative learning, and end-to-end security across IoT, Edge, and Cloud environments.
In parallel, secure and scalable engineering practices were introduced through DevSecOps and MLOps pipelines, enabling continuous integration, testing, and validation. An operational testbed was implemented and validated at TRL5, providing a stable environment for early experimentation and system integration.
On the scientific side, CoGNETs developed the foundations of decentralized, game-based intelligence. Optimization objectives were formalized using utility functions capturing latency, throughput, reliability, security, energy efficiency, and system welfare, and mapped to resource availability. A game-theoretic baseline was established, covering suitable game models, equilibrium concepts, auction mechanisms, and learning strategies for dynamic computing continuums.
Based on this framework, autonomous game-intelligent agents were designed and initially implemented to support on-device decision-making. Selected game-theoretic formulations were translated into agent-based solutions, resulting in initial software components enabling decentralized optimization.
Further work addressed collaborative intelligence through federated and split learning approaches, including distributed training models, neural network optimization techniques, and AI models aligned with the pilot domains. End-to-end security was also assessed, covering identity, privacy, resilience, and trust in heterogeneous environments.
At the middleware level, a programmable infrastructure was designed and partially implemented to orchestrate distributed resources across Edge and Cloud systems, including container-based orchestration, cluster federation, and intelligent workload distribution. Core middleware services—such as identity management, the Cognitive AI Service Repository, and a human-in-the-loop dashboard—were specified and progressively integrated, supported by deployment and operational guidelines.
Finally, CoGNETs initiated the integration of its technologies into pilot domains in manufacturing, connected driving, and healthcare. The testbed was upgraded to support pilot experimentation, initial demonstration and validation plans were prepared, and a common validation framework was defined, including baseline KPI assessments and a high-level evaluation methodology.
Core efforts focused on defining the cognitive computing continuum and translating it into concrete system requirements for the CoGNETs middleware and pilots. Based on expert input and a comprehensive state-of-the-art analysis, the overall CoGNETs architecture was defined, integrating game-theoretic intelligence, collaborative learning, and end-to-end security across IoT, Edge, and Cloud environments.
In parallel, secure and scalable engineering practices were introduced through DevSecOps and MLOps pipelines, enabling continuous integration, testing, and validation. An operational testbed was implemented and validated at TRL5, providing a stable environment for early experimentation and system integration.
On the scientific side, CoGNETs developed the foundations of decentralized, game-based intelligence. Optimization objectives were formalized using utility functions capturing latency, throughput, reliability, security, energy efficiency, and system welfare, and mapped to resource availability. A game-theoretic baseline was established, covering suitable game models, equilibrium concepts, auction mechanisms, and learning strategies for dynamic computing continuums.
Based on this framework, autonomous game-intelligent agents were designed and initially implemented to support on-device decision-making. Selected game-theoretic formulations were translated into agent-based solutions, resulting in initial software components enabling decentralized optimization.
Further work addressed collaborative intelligence through federated and split learning approaches, including distributed training models, neural network optimization techniques, and AI models aligned with the pilot domains. End-to-end security was also assessed, covering identity, privacy, resilience, and trust in heterogeneous environments.
At the middleware level, a programmable infrastructure was designed and partially implemented to orchestrate distributed resources across Edge and Cloud systems, including container-based orchestration, cluster federation, and intelligent workload distribution. Core middleware services—such as identity management, the Cognitive AI Service Repository, and a human-in-the-loop dashboard—were specified and progressively integrated, supported by deployment and operational guidelines.
Finally, CoGNETs initiated the integration of its technologies into pilot domains in manufacturing, connected driving, and healthcare. The testbed was upgraded to support pilot experimentation, initial demonstration and validation plans were prepared, and a common validation framework was defined, including baseline KPI assessments and a high-level evaluation methodology.
The results beyond the state of the art will be completed in the final report.