with Managed Federated Machine Learning
prenode’s solution enables you to enrich your products with powerful AI-based features and services without sacrificing data privacy. Incorporating AI in your existing offerings can increase customer satisfaction, generate additional revenue and strengthen your market position.
Machinery failure needs to be predicted in order to offer maintenance service before an actual malfunction. Each machine type is used by several customers, each with an individual configuration. Historical data per customer offers only a few observations of malfunctions and is not sufficient to develop a prediction model.
Historic and real-time sensor data of each customer is collected by the IoT solution in a private-cloud environment. Using mlx, machine learning models are developed based on all customer data. The deployed predictive model is individualized towards the environment and the machine configuration of each customer.
mlx leverages the sensor data of all customers of the machine manufacturer using Federated Machine Learning. The manufacturer can use a well performing predictive model for machinery failure despite a small number of observations per customer.
Each machine is slightly different, and each customer operates in an individual environment. The predictive models are trained on all available sensor data across customers but are individualized for every customer to ensure the best possible performance.
Model performance can be monitored, and models can be updated regularly.
Franchise-restaurant chains with multiple branches need to anticipate future sales in order to purchase and stock food supply as well as plan required staff. Historic sales data of one restaurant is not enough to create well-performing forecasting models.
Solution provider offers a software-as-a-service solution and stores all historic sales data in separate databases in a private cloud. mlx accesses the non-sensitive historical sales data and develops forecast models using its unique Federated Transfer Learning approach. Sales forecasting models are deployed in the cloud of the solution provider to predict future sales as a feature of the gastronomy software .
Participating restaurants can forecast sales better and as a result optimize supply and personnel planning. This sales forecast model for daily revenues helped decrease the individual model mean average percentage error (MAPE) by up to 34.2%.
mlx ensures that individual customer data is not shared with any other customer. In addition, using different security mechanisms, mlx also guarantees that the developed predictive model does not reveal the data it was trained on.
Restaurants without past sales data can use pre-trained sales forecast models.
An ambient system needs to analyze activity patterns based on individual elderly motion data and detect emergencies. Without direct access to motion data, caretakers need to be notified in case of emergencies and doctors need to recognize drifts in behavioural trends.
Patients’ motion data is collected by wrist-bands with motion sensors and stored in local gateways. mlx is used to develop a machine learning classifier on the local gateway. A machine learning model individualized towards each patient is deployed to detect individual activity patterns.
Healthcare system provider can offer its customers a new AI-based services that allow elderly to stay at home longer and caretakers are reliably notified if needed.
Patient data is not transferred or shared with anyone.
The healthcare provider can use pre-trained machine learning models for new patients and thus use the system from day one.