Studio is a comprehensive cloud-native platform for secure and data-private machine learning. Built on MLSecOps principles, Studio is modular and lets a customer establish a tailored, secure on-premise machine learning environment where models can be trained and deployed with full control over data-and model access. In this way an organization can enable an intranet of models including both home-grown models and generalist models such as LLMs. Federated Learning is a first-class citizen in Studio, providing the capability to fine-tune models on distributed private data. Scaleout Studio is available as a subscription for on-premise installations, and for trials as a SaaS.
FEDn is an open framework for highly scalable federated learning. The FEDn SDK allows development of custom federated learning solutions and integrations. FEDn is available under a dual license, as FOSS under Apache2 with community support, and under a Scaleout commercial license including Enterprise support.
The need for data-private machine learning solutions in the Enterprise
The rapid progress in machine learning has been evident with models like ChatGPT and Midjourney, which are examples of generalist models trained on large amounts of public data. However, when machine learning is applied to specific industrial applications, models need to be trained or fine-tuned on data specific to the application. Also, there are many use-cases for which it is not possible to send data to an external AI service.
At Scaleout we saw the complexity involved in deploying and supporting machine learning environments on-premise to leverage local data. There are powerful open source stacks for machine learning operations, but it is not easy to make use of them, and a lot of the complexity around security and access control is left to the end-user. With Studio we have created a modular framework that makes it easy to develop and deploy models in your own secure environment, and that can grow with functionality as your needs evolve.
However, it is not enough with an on-premise AI stack. Currently, challenges arise around data access. Often, access to data is limited for various reasons, which in turn limits the quality of models that can be created. Examples of this include geographically dispersed data such as patient data from different hospitals, process data from different factories in different countries, user-generated data on mobile phones, or sensor and image data generated directly on vehicles, to name a few contexts.
It is often difficult to collect this data in one place, such as in a central cloud infrastructure. Data security, regulatory requirements (GDPR), and difficulties and costs associated with moving large amounts of data ("big data") are today a limitation for industrial application of machine learning. At the same time, a strong focus on data security and privacy is essential for the future safe use of AI. This is why we have built federated machine learning - a form of decentralized machine learning - into the DNA of our technology stack.
"At Scaleout, we tackle distributed data with software that allows us to move machine learning to the data instead of moving the data to machine learning. In this way, models can learn from distributed data, but we avoid moving and thereby exposing data”, explains Andreas Helllander, CEO and co-founder at Scaleout.
Scaleout's solution and technology has already helped a number of enterprises, including Scania, SciLifeLab, Eurocontrol, SITA Aero, and Swiss airlines, to address data access challenges and to securely train machine learning models on data from multiple sources, without the need to centralize that data in a single location.
We are excited to partner with investors ALMI Invest, Beijer Ventures, Uppsala University Invest, and technology entrepreneurs and angel investors Michael Lantz (Accedo Broadband) and Fredric Wallsten (Safespring). Together we support a future where ML solutions become the best they can be, but without having to compromise with data ownership and privacy.