Practical example and current challenges
- Most companies have already recognized that data has enormous value. They want to differentiate themselves from their competitors by offering their customers tailored digital services. At the same time, car manufacturers are tasked with optimizing the energy efficiency and consumption of electric vehicles – which could be improved by using Machine Learning (ML) models. These models need to be trained with the use of in-vehicle data. The models, as well as the data, however, are extremely valuable and represent a potential differentiator for the party controlling it. Therefore, not every company is willing to share or even sell data. Instead, companies are emphasizing the importance of data privacy, data sovereignty, and data ownership. Not only is data provenance important, but also local retention of data. The reason is that some data points are too big to be shared, while other data points are too sensitive to be shared.
- This challenge can be solved by decentralized learning. Multiple parties train ML models together without exposing their data. Instead, they train the model locally using their own data, and then share the insights from the current training cycle. This process needs to be orchestrated. A solution for these challenges is proposed, that introduces required management functions and enables participating organizations to join and leave swarms, submit ML models, and collaborate in an ML ecosystem. It provides a data-centric solution for rapid contracting, real-time contract monitoring, and trusted and secure digital service exchange that enables an open ecosystem for decentralized learning.
- The described AI-system is particularly applicable in situations where car manufacturers, tiered suppliers, battery manufacturers and recycling or charging station operators train ML models locally in-vehicle to optimize electric vehicles’ energy efficiency and consumption. Additionally, decentralized ML could be applied in the area of predictive maintenance. In this way, in-vehicle sensor data is used together with historic data to combine central and decentralized ML for predicting life cycle events (tear and wear) of parts and components of the vehicles. This results in continuous quality improvements of parts and components. The approach also enables real-time control and simulation by leveraging ML directly in-vehicle and on real-time data, and not siloed to a single vehicle.
- A main advantage is the open source, platform-agnostic and completely decentralized nature of the system, which means that there is no central custodian accessing the participants' data or models. Thus, decentralized ML enabled by the solution is handled directly between the model providers and the consumers, without the need to involve any third party. The solution also avoids lengthy negotiations as consumers can join swarms with the click of a button and thus enables an automated onboarding. The AI-system also follows the Gaia-X principles by ensuring data sovereignty: Data privacy and protection is ensured, ownership is respected and protected throughout the ecosystem and each party can decide with whom to interact.
What added value does the "Gaia-X project" offer?
- Many companies and research institutions are working on several projects in the area of "Indus-try 4.0 / Smart Manufacturing / Artificial Intelligence in Manufacturing". This is certainly associated with a tremendous amount of research funding. Gaia-X would be a considerable acceleration and the solution to the same problem: data sovereign access to data and its distribution, compatible and neutral.
- Topics about data ownership, privacy, and sovereignty, but also technical, functional, regulatory, and legislative limitations keep companies from connecting and collaborating. Gaia-X helps to overcome the hurdle of trust for microservice architectures by decentralizing computing power and data across traditional boundaries.
- Gaia-X enables new technologies like decentralized ML and allows new business models based on sharing data to leverage data value and to (re-)gain competitive advantage.
Use Case Team
- Christine Wenzel – Hewlett Packard Enterprise
- Thomas Ernst – Hewlett Packard Enterprise
- Hartmut Schultze – Hewlett Packard Enterprise
- Florian Bühr – Hewlett Packard Enterprise
- Marcus Friedrich – Hewlett Packard Enterprise