Periodic Reporting for period 2 - AI-SPRINT (Artificial Intelligence in Secure PRIvacy-preserving computing coNTinuum)
Período documentado: 2022-07-01 hasta 2023-12-31
The AI market size is projected to grow to USD 2,025.12 billion by 2030. Many experts believe the innovation will come from computing resources at the periphery of the network. Many companies are evaluating the use of edge computing for data collection, processing, and online analytics to reduce application latency and data transfers.
The focus of AI-SPRINT is to define a novel framework for developing and operating AI applications, together with their data, exploiting computing continuum environments, which include resources from the edge up to the cloud. AI-SPRINT offers novel tools for AI applications development, secure execution, easy deployment, as well as runtime management and optimization. AI-SPRINT tools allow trading-off application performance, energy efficiency, and AI models accuracy while providing security and privacy guarantees. The AI-SPRINT framework supports AI applications data protection, architecture enhancement, agile delivery, runtime optimization, and continuous adaptation. AI-SPRINT makes it possible to seamlessly design and partition AI applications among the many cloud-based solutions and AI-based sensor devices.
The following objectives were achived:
- Provide novel design tools for AI applications: development tools for the implementation of AI applications consuming resources across the computing continuum including multi-clouds, edge servers, and AI-enabled sensors.
- Deliver tools for secure execution and privacy preservation: Defined solutions providing secure AI models deployment and data processing that combine hardware-enforced confidentiality and integrity with policy-defined privacy.
- Runtime environment for application execution and monitoring: Offer a runtime environment and management policies orchestrating the applications execution across the computing continuum with performance guarantees.
- Advanced solutions for AI architecture enhancement: Defined an approach and solutions to iteratively refine AI application architecture design and deployment.
- AI framework validated through use cases: AI-SPRINT solutions have been validated across industrial and health domains through three use cases including farming 4.0 maintenance and inspection, and personalized healthcare in line with EU priority areas for AI investments.
The focus of AI-SPRINT is to define a novel framework for developing and operating AI applications, together with their data, exploiting computing continuum environments, which include resources from the edge up to the cloud. AI-SPRINT offers novel tools for AI applications development, secure execution, easy deployment, as well as runtime management and optimization. AI-SPRINT tools allow trading-off application performance, energy efficiency, and AI models accuracy while providing security and privacy guarantees. The AI-SPRINT framework supports AI applications data protection, architecture enhancement, agile delivery, runtime optimization, and continuous adaptation. AI-SPRINT makes it possible to seamlessly design and partition AI applications among the many cloud-based solutions and AI-based sensor devices.
The following objectives were achived:
- Provide novel design tools for AI applications: development tools for the implementation of AI applications consuming resources across the computing continuum including multi-clouds, edge servers, and AI-enabled sensors.
- Deliver tools for secure execution and privacy preservation: Defined solutions providing secure AI models deployment and data processing that combine hardware-enforced confidentiality and integrity with policy-defined privacy.
- Runtime environment for application execution and monitoring: Offer a runtime environment and management policies orchestrating the applications execution across the computing continuum with performance guarantees.
- Advanced solutions for AI architecture enhancement: Defined an approach and solutions to iteratively refine AI application architecture design and deployment.
- AI framework validated through use cases: AI-SPRINT solutions have been validated across industrial and health domains through three use cases including farming 4.0 maintenance and inspection, and personalized healthcare in line with EU priority areas for AI investments.
In the 36 months of activity of the project, all foreseen activities have been performed and all deliverables released. In RP2, the following stand out: 1) the development of the final release of the core tools for the design, deployment, and secure execution of AI applications, 2) the final release of use case applications (stroke assessment, maintenance & inspection, and farming 4.0) with the final evaluation and validation of the design, runtime, and security tools. WP1 worked on finalizing the definition of the AI-SPRINT reference architecture according to the homogenisation/integration performed for the core tools according to the development of new tools, fulfilling its corresponding objectives. WP2, WP3, and WP4 delivered the final release of the core tools for the design, deployment, runtime and secure execution of AI applications. WP5 delivered both the consolidated version of the use case application components with the evaluation and assessment of the core tools, while WP6 consolidated the AI-SPRINT Alliance and finalized the exploitation strategy. Partners kept the fruitful collaboration through the biweekly and WP-focused virtual meetings. Additionally, physical meetings were also organised.
Among the key activities in relation to result dissemination performed during RP2, there was the active participation at technical and business venues, as well as in conferences and events, which ensured that AI-SPRINT findings reached both technical and non-technical audiences, fostering a broader impact. The establishment of successful synergies and collaborations with relevant initiatives in the fields of AI, Edge Computing, and Open Source Software contributed to a more extensive reach and enhanced the project influence.
The exploitation activities focused on the identification of 14 exploitable assets tailored to specific characteristics and stakeholders and addressing distinct challenges and opportunities in the AI landscape. A set of four asset combinations for joint exploitation activities among selected partners have been set up. Specific Exploitation Agreements were established, defining the organisational model and revenue sharing between the involved partners. The emphasis on joint exploitation underscores the commitment towards a shared vision of leveraging AI-SPRINT assets.
An essential achievement was the steady growth of a robust community comprising diverse stakeholders, including SMEs, Cloud providers, and members from the Research and Academia and guiding partners in effective exploitation planning and alliance management. A start-up related to the healthcare use case will be founded by the BSC partner.
Among the key activities in relation to result dissemination performed during RP2, there was the active participation at technical and business venues, as well as in conferences and events, which ensured that AI-SPRINT findings reached both technical and non-technical audiences, fostering a broader impact. The establishment of successful synergies and collaborations with relevant initiatives in the fields of AI, Edge Computing, and Open Source Software contributed to a more extensive reach and enhanced the project influence.
The exploitation activities focused on the identification of 14 exploitable assets tailored to specific characteristics and stakeholders and addressing distinct challenges and opportunities in the AI landscape. A set of four asset combinations for joint exploitation activities among selected partners have been set up. Specific Exploitation Agreements were established, defining the organisational model and revenue sharing between the involved partners. The emphasis on joint exploitation underscores the commitment towards a shared vision of leveraging AI-SPRINT assets.
An essential achievement was the steady growth of a robust community comprising diverse stakeholders, including SMEs, Cloud providers, and members from the Research and Academia and guiding partners in effective exploitation planning and alliance management. A start-up related to the healthcare use case will be founded by the BSC partner.
AI-SPRINT extended the state of the art in terms of design and programming abstraction, performance models and applications design space exploration, and AI models architecture search.
The AI-SPRINT framework is more flexible than existing programming solutions. AI-SPRINT addresses most of the challenges related to the composition of AI applications on computing continua by introducing simple code annotations. The applicaiton can then be automatically deployed and executed on the edge with transparent offload to cloud platforms depending on load and constraints requirements. Specific highlights have been put on security and QoS in the definition of the framework. The COMPSs programming model has been used as base technology and extended to support live migration and energy aware execution of long running jobs. Applications can be designed transparently with respect to the AI framework and their execution is delegated to OSCAR according to a FaaS model.
ML-based performance models have been developed to predict the execution time of applications running on heterogeneous multi-devices systems composed of IoT & AI enabled sensors and edge and cloud resources. Techniques to identify the best computing continuum configuration while satisfying all the QoS constraints have been implemented within the SPACE4AI-D tool. Design time solutions are exploited by the runtime reconfiguration algorithms implemented by SPACE4AI-R.
In AI-SPRINT, architecture search considers multiple trade-offs, i.e. AI models accuracy, performance, energy consumption, and data privacy. Both deep networks for image processing and time series analysis are supported.
The AI-SPRINT tools have been validated by the three Use Cases related to farming 4.0 mantainance & inspection of wind farms and personalised healthcare addressing the following SDGs:
- 3 good health and well-being
- 5 gender equality (gender stroke disparities of the heatlhcare UC)
- 6 clean water and sanitization (reduction of phytosanitaris in farming UC)
- 7 affordable and clean energy (development of AI models for inspecting wind farms)
- 9 industry, innovation and infrastructure
- 12 responsible consumption and production (energy efficiency solutions)
- 13 climate action (improvement in farming practices contributes to reduction in climate change)
- 15 life on land (reduction of used pesticides and lowering the ecological footprint of energy -wind- and infrastructure systems)
The AI-SPRINT framework is more flexible than existing programming solutions. AI-SPRINT addresses most of the challenges related to the composition of AI applications on computing continua by introducing simple code annotations. The applicaiton can then be automatically deployed and executed on the edge with transparent offload to cloud platforms depending on load and constraints requirements. Specific highlights have been put on security and QoS in the definition of the framework. The COMPSs programming model has been used as base technology and extended to support live migration and energy aware execution of long running jobs. Applications can be designed transparently with respect to the AI framework and their execution is delegated to OSCAR according to a FaaS model.
ML-based performance models have been developed to predict the execution time of applications running on heterogeneous multi-devices systems composed of IoT & AI enabled sensors and edge and cloud resources. Techniques to identify the best computing continuum configuration while satisfying all the QoS constraints have been implemented within the SPACE4AI-D tool. Design time solutions are exploited by the runtime reconfiguration algorithms implemented by SPACE4AI-R.
In AI-SPRINT, architecture search considers multiple trade-offs, i.e. AI models accuracy, performance, energy consumption, and data privacy. Both deep networks for image processing and time series analysis are supported.
The AI-SPRINT tools have been validated by the three Use Cases related to farming 4.0 mantainance & inspection of wind farms and personalised healthcare addressing the following SDGs:
- 3 good health and well-being
- 5 gender equality (gender stroke disparities of the heatlhcare UC)
- 6 clean water and sanitization (reduction of phytosanitaris in farming UC)
- 7 affordable and clean energy (development of AI models for inspecting wind farms)
- 9 industry, innovation and infrastructure
- 12 responsible consumption and production (energy efficiency solutions)
- 13 climate action (improvement in farming practices contributes to reduction in climate change)
- 15 life on land (reduction of used pesticides and lowering the ecological footprint of energy -wind- and infrastructure systems)