Features of Katulu Federated Learning (FL)

Automatic clustering for reduced complexity

Machines are used in different environments and processes and are designed accordingly. Katulu FL automatically detects which machines and assemblies can be trained together according to their design and automatically groups them into clusters that can be trained together. There is no need to determine in advance which machines and designs can learn from each other. Changes in the design of a machine are automatically detected and lead to an adjustment of the clustering. This ensures that each machine - no matter how individual - learns from other comparable machines or assemblies.

Federated Learning for Special Purpose Machines

Special purpose machines (SPM) are unique - and so is their data. This is why Katulu focuses on training models for standardized assemblies. By combining these models, meaningful models can be mapped for each individual SPM.

Privacy by Design

Privacy protection is firmly embedded in the core of Katulu FL. The use of Differential Privacy completely prevents possible inferences from the model to any of the industrial users. Homomorphic Encryption also allows encrypted data to be processed without the need for decryption.

Asynchronous & Offline Communication

Katulu FL does not require a permanent connection to the cloud. The decentralized solution at the customer site runs entirely on the edge - independent of the cloud. Whether temporary disconnections or controlled offline switching of a system, this has no impact on the functioning of the models and the multi-defense services they use. A connection to the cloud is only required to synchronize the models. A physical data transport solution is also possible on request if guidelines or technical circumstances do not allow a connection to the cloud.

Seamless integration with existing infrastructure

As a platform-agnostic solution, Katulu FL is based on Kubernetes and supports all cloud providers (AWS, Azure, Google, IBM, IONOS, GaiaX, etc.). Standardized interfaces are available for integration, allowing the models to be used in other systems like IoT platform, MES, SCADA, ERP, etc., OPC-UA and MQTT are supported as industry protocols for data acquisition - other protocols are possible on request.

Katulu Edge AI Box

The Katulu Edge AI Box is an edge gateway specifically designed for the requirements of decentralized machine learning in industry. The high-performance and energy-efficient Nvidia Jetson TX2i module allows use in extreme temperature ranges from -20°C to +70°C due to its passive cooling. The availability lifecycle of this industrial variant is 10 years.

Familiar Data Science Tools

No experimentation! With Katulu FL, Data Scientists continue to work with familiar tools such as Tensorflow, Keras, Jupyter, and PyTorch.