Apheris Compute Gateway - Let's collaborate.

Collaboration is essential for our survival and success as a society. In today's data-driven landscape, this principle is crucial for tackling the world's most pressing challenges. Yet, safeguarding sensitive information is paramount.

Today we are very excited to announce the general availability of the Apheris Compute Gateway 3.0. With this third major release, the Apheris team provides the first end-to-end Computational Governance Solution for federated statistics and machine learning to the market - enabling everyone to safely collaborate even on highly sensitive data.

A lot has happened over the past year, most importantly a huge amount of hard, dedicated work of the whole team in close collaboration with our exceptional customers, partners and research institutions. Thank you all so much for your patience, ideas, feedback and collaborative spirit!

What's new?

Core to our offering has always been the Compute Gateway - a lightweight software agent easily deployed on any infrastructure and running within a data provider’s environment. It enables oversight of computations on the algorithmic level and ensures data stays always behind a data custodian’s firewall. The Compute Gateway will only return analytical results such as model weights and never raw data.

The Compute Gateway has had a substantial upgrade - more on that in just a bit - and is embedded within a federated architecture.

The new Federation Engine

About a year ago, the Apheris team decided to implement NVIDIA FLARE as our new federation engine. FLARE is battle-tested in many high-risk projects and supplies the Apheris solution with a reliable, scalable, fast, adaptive and secure backbone for federation. Changing a fundamental component like a federation engine is no easy endeavor but the exceptional team behind FLARE has been of tremendous help and inspiring to collaborate with. Thank you all for your contributions!

ML Engineers now have a simple path to the worldwide largest collaborative community for ML models, such as Hugging Face and NVIDIA Catalog, as well as the flexibility to define their own custom models. Any data-driven algorithm, from statistics right up to deep learning, is ready for federation out-of-the-box.

The Governance Portal - Control everything in one place

Once the Compute Gateway is deployed, users can control everything via the newly improved UI, which allows the Data Custodian to:

• Register and manage datasets for the Gateway (sensitive data never leaves the environment)

• Control computations via Asset Policy configuration (not compliant = not executed)

• Review and approve computations via Compute Specs and access detailed information about used models via the Model Registry (Trust is good. Control is better.)

• Trace and log any interaction with data for systematic auditability

Everything handy in one place.

A Computational Handshake

Via federation, the Compute Gateway allows data to stay where it is and bring computations to the data. But in many cases, this is not enough to comply with internal policies or requirements set by regulators. What you want is a way to operationalize these requirements and enforce them automatically on an algorithmic level. With Asset Policies and Compute Specs, data custodians and ML engineers can realize this computational handshake.

The new Apheris workflow consists of 3 simple steps:

Prepare: Data Custodians define Asset Policies and ML Engineers specify computations (Compute Specs)

Agree: After the ML Engineer submitted a Compute Spec the Compute Gateway validates automatically and provides a review option for the Data Custodian for final approval

Use: Once a Compute Spec has been approved, compute jobs can be run which are always controlled by the Compute Gateway to ensure the agreed specifications are met

This workflow allows even the most sensitive datasets can be contributed to research in a secure, privacy-preserving and compliant way.

Let's Collaborate!

Integrating NVIDIA FLARE as our federation engine has significantly enhanced product functionality and facilitates collaboration with other major Open Source communities.

The Apheris Compute Gateway allows AI models to be safely developed on distributed data. The simple process of prepare - agree - use of compute specifications in a streamlined workflow helps ensure that computations adhere to strict privacy and security standards. This setup allows Data Scientists or ML Engineers to validate data-driven workflows before submission, which reduces friction, while Data Custodians stay in full control of what is being done with their data.

Furthermore, the Compute Gateway's compatibility with any data type and model, along with its comprehensive approach to privacy, security, and governance, establishes Apheris as a front-runner in secure, federated computing.

“Safeguarding sensitive hospital data is critical to ensure patient privacy and data security,” said Holger Roth, Principal Federated Learning Scientist, NVIDIA. “Integrating NVIDIA FLARE with the Compute Gateway 3.0’s governance, security and privacy capabilities unlocks new possibilities to make a difference in healthcare and beyond.”