OpenFLaaS
Open Federated Learning as a Service
| Short description |
| The OpenFLaaS project is developing an AI framework for the automated, decentralised analysis of industrial documents such as tenders, contracts or product data sheets in order to digitalise knowledge in companies efficiently and securely. |
OpenFLaaS is creating a modular system for AI services that companies can use for automatic analysis of documents such as tenders, contracts and product data sheets, thus helping them to make efficient use of the knowledge contained in these documents in digitised form. With the OpenFLaaS platform, the AI services are executed in protected spaces on end devices (edge devices) in the company. The raw data from the documents remains in the company. Secure, sovereign and trustworthy exchange of processed data is combined with the AI deployment across various devices. Both processes take place in data spaces that comply with the Gaia-X standard. With this approach, OpenFlaas is promoting the secure and efficient digitalisation of corporate knowledge to enhance productivity while preserving the data sovereignty of companies. OpenFlaaS is simultaneously providing a foundation for the development of a Gaia-X-compatible ecosystem for intelligent document analysis (IDA) and the development of system-independent, innovative document analysis services.
Market perspectives and product claims
Companies collect their knowledge, their core corporate asset, in countless documents. With the help of AI, this knowledge can be analysed to increase efficiency, identify new correlations and thus extract new information, support decision-making, expand service offerings and much more. Given these advantages, it is not surprising that even in its relative infancy, AI-supported document analysis already has a global market volume of over one billion US dollars.
However, an obstacle currently hindering IDA is the fact that computationally intensive analyses are performed on cloud computers. Many companies are reluctant to fully harness the potential of IDA, due to data protection reasons and a lack of trust in cloud providers. The OpenFLaaS project resolves this issue by offering companies the option to use IDA as a service (aaS), in which case the data remains within the company. This is enabled by the development of edge AI spaces, which are simply connected to Gaia-X-compliant data spaces via plug-and-play. OpenFLaaS thus offers companies a cloud edge system for intelligent document and text analysis services on edge devices such as industrial systems and machines, computers, smartphones or tablets. Companies retain full control over their data at all times when using these services.
Based on the OpenFLaaS project results, a myriad of IDA-based value-added products and services can be developed in Germany and Europe. These products and services can in turn be continuously improved using the project solution, both in terms of their general application to any area of knowledge and their adaptation or specialisation with regard to specific domains. Thanks to innovative algorithms, the CO2 emissions of the analysis processes are minimised since fewer data transactions are required between the edge and the cloud. Legal discussions and court rulings in the area of data use for text and data mining (TDM) are still in flux. The project will assess the legal risks in this area and define the framework conditions under which the data products can be made available for commercial use.
Challenge and innovation
OpenFLaaS focuses on the decentralisation of AI for intelligent document analysis. Document analysis takes place not in a central cloud, but across edge devices. The foundation for this approach is federated learning, a distributed machine learning process in which only a global AI model is trained in the cloud, while the training data is stored locally, thus ensuring that solutions comply with data protection regulations.
The innovation led by OpenFlaas is to combine the principles of federated learning at the edge level within protected AI spaces with the open standards of Gaia-X. This combination enables the training of AI models for intelligent document analysis while preserving data sovereignty. The project will develop and implement secure, federated data spaces in a cloud-edge continuum that complies with Gaia-X. Using the example of industrial document types such as contracts, technical CAD drawings, and tender specifications, OpenFLaaS is developing federated intelligent services that are capable of understanding and analysing different text types and layouts such as tables, body text, headings, formulas, images, headers and footers, etc. The software developed in the project is released as open source.
Use cases
The project innovation will be demonstrated using a prototype for the planning and tendering of industrial plants, based on real data. In a second demonstrator, the same principles for AI-assisted analysis of clinical documents about patients will be transferred and implemented on edge devices. Detailed requirements from operators and users regarding technical implementation, data sovereignty, and user-friendliness are collated. The prototype applications are then validated by experts from the two application fields.
Consortium
DRIMCO GmbH (consortium leader), Apheris GmbH, Siemens AG, Fraunhofer IAIS, Goethe University Frankfurt, Institute for Applied Informatics, Leipzig (InfAI); University of Vienna (Austria)
Benefits
| Current situation | Future vision |
| For industrial document analysis, companies in Germany and Europe are largely dependent on American and Chinese cloud providers, who hold a monopoly or oligopoly position. Due to data protection concerns and a lack of trust relating to these providers, IDA is often not used at all or its full potential remains untapped. | OpenFLaaS facilitates decentralised execution of IDA at the edge level, eliminating the need for companies to upload data to the cloud. This approach alleviates concerns about potential loss of knowledge through data and ensures compliance with data protection regulations, even when dealing third parties. |
| Due to the lack of trustworthy, secure spaces for the exchange of data, the development of new digital business models and services in the IDA sector lags behind in terms of the technological options available. | OpenFLaaS creates secure and trustworthy AI spaces for the exchange of data, thus enabling new digital business models and services for an edge-oriented data economy. |
| In industry, there are many specific document types containing specific industry knowledge. AI requires special training to analyse this knowledge. Currently there are no services offering the training and development of such highly specialised IDA. | By facilitating the establishment of a Gaia-X-compatible node for industrial document analysis in Germany, OpenFLaaS is helping to promote the development of services for AI-based industrial decision support aimed at optimising productivity. |
| Established providers in the IDA sector offer solutions that often operate as “black boxes”, preventing customers from understanding the actual workings of the services. | OpenFLaaS creates transparency about IDA-based value-added products and services through an open-source solution. In this way, the project builds trust in automated document analysis. |
| AI-based document analysis is associated with high data transfer rates to the cloud and therefore causes high CO2 emissions. | OpenFLaaS enables sustainable IDA since its solution transfers less data to the cloud. |
