A lot has happened at Apheris over the past few months: The whole team is excited about the new collaboration announcements with AI Structural Biology Consortium, NVIDIA, AWS and bringing the Compute Gateway also to the AWS Marketplace. And because our engineering team is a dedicated bunch of awesomeness, we added many important features and improvements to our product at the same time.
With the release of Apheris 3.2 today, you can benefit from many improvements, especially to the Data Science and ML Engineer user experience.
Apheris 3.2 highlights include:
Extended capabilities: Two new models got added to the Apheris Model Registry and are, of course, federation-ready (details below)
Apheris CLI – lighter and faster
Simplified Compute Specs
Improved error handling and messages
Extended documentation and simplified learning
Added expert content to the Trust Center, including the executive summary of our latest penetration tests
If you are already familiar with Apheris, you know that we strive for simplicity when enabling secure, governed and privacy-preserving federation. As simple as a handshake, just computational.
New Models added
We don’t force our models into high heels but they do live in the spotlight… as well as our Model Registry.
Models in the Apheris Model Registry are pre-ported, federation-ready ML and Data Science algorithms ready for use out-of-the-box. We added two new models extending the analytical capabilities you’ll get with Apheris and they easily integrate into your data science workflows.
Federated nnU-Net
An autoML approach to 2D and 3D medical image segmentation.
Uses a privacy preserving fingerprinting method to analyze remote datasets at each site, aggregate it and use the aggregated fingerprint to determine the model structure and hyperparameters.
Both federated training and inference are supported on Apheris.
Allows out-of-the-box segmentation of image, nifti and 3D tiff files.
Federated Cox Regression
Part of the Apheris Regression Models repository
Determine the effects of several variables on the time to event
Based on the popular Lifelines package, extended to support federated training.
We will continuously add models to the Model Registry and, of course, you can always add your own models to your organizations deployments. As Apheris uses NVIDIA FLARE under the hood, you can comparatively easy port models from commonly used model libraries such as Hugging Face or BioNeMo.
By the way, our team is hiring a Senior ML Engineer and a Senior ML Field Engineer. If you want to allow the world to safely collaborate on data, join us!
Improved data science and ML experience
The famous chef Marco Pierre White likes to say “Perfection is many things done well” and our team strives for this perfection step-by-step alongside the feedback of our customers and partners. Thank you so much for helping us improve the Apheris experience for everyone.
A few milestones we already achieved with Apheris 3.2:
Simplified the compute spec creation / activation flow
Improved performance of the CLI by removing heavy packages and optimizing imports
Improved CLI error handling to increase transparency and interpretability of errors
Better aligned terminology across the Apheris product
Refactored statistics code architecture to improve encapsulation and reduce redundancy between packages
A big "thank you!" goes out to the entire Apheris team for your dedication to customers, your pursuit of excellence, and for making this great next step possible
