The AIDE project will implement an integrated Dissemination, Exploitation, and Communication (DEC) plan to bring the project’s results to the attention of early adopters and stakeholders. The DEC plan will maximize the project’s impact and, in particular, lay the foundations for future exploitation of the project’s results. Our team’s engagement will help translate research results and developments to the various players in the target industries (suppliers, customers, end-users). Our objectives comprise:
(1) to assess the impact of federated learning on selected industrial cases and
(2) to create avenues for exploiting AIDE’s results.
In terms of impact, several avenues have been identified:
- Provide technology components and open source tools for federated AI that align with relevant EU policy and legal and ethical requirements, enabling application in a wide range of domains (cybersecurity, IoT being specifically targeted in this project). The federated platform features verification mechanisms for sharing local models and deploying global models in compliance with fairness, confidentiality and ethical requirements. Thanks to AIDE’s meta-learning phase, new models for selected use cases are continuously learned and thus keep pace with changes in the federated subsystems. Policy-driven data protection will simplify data’s ethical and privacy-friendly use, as required by the GDPR and the EU’s ethical guidelines for trustworthy AI.
- For the cybersecurity case studied, AIDE’s federated learning platform and AI-based services will help strengthen cybersecurity protections in complex distributed systems and automate various tasks in an organization’s cybersecurity risk management phases [NIST18].
- In the case of the IoT, AIDE’s federated learning platform will enable industrial participants to share knowledge with a global model in a secure, privacy-protected way, benefiting in turn from the more precise model assimilated by the contributions of the various participants (in terms of predictive maintenance, for example).
- A better understanding of how an attacker might use AI technology to attack IT systems. The AIDE platform and cybersecurity services adapt quickly to meet new threats and maintain protection levels. Using federated and meta-learning ensures that AI-based malicious activities designed to evade detection can be detected on a distributed basis by correlating data from multiple subsystems. Meta-learning ensures that countermeasures to new threats are rapidly developed, implemented and deployed to protect the distributed system. Individual AI-based cybersecurity services are designed to be deployed on a federated learning platform and can use the underlying features of federated learning to detect AI-based adversary activity.
- Validation opportunities (from WP1 and WP3) at each scale. In a pilot phase, federated learning can be tested on a single partner’s production site, integrating data and models. At this stage, validation of the effectiveness of federated learning can be assessed even if privacy safeguards still need to be (fully) implemented. If the outcome of a single-site, multi-site pilot case gives satisfactory results, then follow-up can be extended to multi-site operation. AIDE’s real-world resilience against AI-based attacks on federated systems (WP2) can be tested during multi-site operations. If successful, the project can invite new industrial members to join (WP5), and privacy aspects will be fully considered. Private, federated learning enables several partners to benefit from global predictive maintenance models without having to share data or operational details.
Longer-term impacts have also been identified:
- Respecting privacy and other fundamental rights: Privacy and fairness are essential conditions of today’s technology. Without them, trust and acceptance will be limited. By introducing strong privacy and fairness guarantees for deployed AI, we increase impact as AI can be deployed in more cases and on more data types. Transparency allows users to understand how their data is used and what benefits accrue.
- Unleashing the new potential for collaboration between industrial parties: Companies are sometimes reluctant to collaborate on AI projects due to the risk of opening up sensitive data to other parties (including competitors). With federated learning, these risks can be minimized, enabling companies to collaborate to create more robust AI applications.
- Increased cybersecurity and a more secure online environment: As systems and systems of systems become increasingly connected, it becomes ever more critical to secure the entire distributed system, not just individual sub-systems. The dependency is such that in these large systems, an intrusion at any point in the system can compromise any part of the system. However, complex distributed systems can be geographically distributed across different sites and organizations, making it challenging to share security-related information. In the cybersecurity use case, the AIDE federated learning platform aims to provide a solution for securing large ecosystems of services that depend on each other.
- Resilient and adaptive protection of software systems to resist and counter cyberattacks and hybrid threats: The security of complex distributed systems must be prepared, respond to, and recover to be resilient to cyberattacks and able to adapt to maintain the level of protection for new threats. The federated learning nature of the platform and its AI-based services is designed to resist AI-based tampering and use adversarial machine learning to deceive defences.
- Generating more robust IoT models for preventive maintenance: Enabling companies embracing Industry 4.0 to share information federally and in complete confidence (without implications for the company’s sensitive data on its products or product lines) leads to the generation of more robust models (for example, in terms of preventive maintenance) and becomes a win-win situation for all parties involved.
In addition, we will use traditional dissemination vectors such as publications and ecosystem animation (see partner ecosystems). We will organize an event each year to disseminate results and interact with potential new players. AIDE members also interact frequently with peers through conferences, collaborative papers, and other professional and scholarly networking means.
| Key Performance Indicators (KPI) | Leader | Contributor | Chronology |
|---|---|---|---|
| ● DEC publication plan ● Publication of the completion of the DEC plan ● List of publications ● Equipment users ● International collaboration ● List of events. | UCLOUVAIN | All | ● End 2023 : first report on dissemination and impact activities In case of extension: ● End 2024 (if extended): second report on outreach and impact activities ● End 2025 (if extended): third report on dissemination and impact activities. |