What's new in the Genius Darwin release?
Description of updates to Genius as of May, 2025
Largely driven by feedback from users of the Curie release (Winter 2025), the Darwin release of Genius (Spring 2025) packages several fundamental functionality additions as well as significant improvements to the usability of the model editor UI. The sections below details the changes made available in Genius during this release.
UI Overhaul
The entire user interface (UI) for the Genius model editor has been overhauled and modified to produce a streamlined and user-friendly experience. This work was undertaken as a result of customer feedback indicating the previous UI often behaved in unexpected ways and created challenges in creating probabilistic models.
Many changes have been made, and you can find out more about using this new UI in the Genius model editor page.
Data-to-Model Wizard Experience
A common challenge faced when using machine learning tools relates to converting raw data into a useful model. To simplify this process and make it easy for anyone to get a powerful model out of their observation data, the Genius model editor now includes a wizard that walks you through the process and automates some of the most tedious steps. This wizard currently accounts for the most common and simple model types, and enhancements are planned to support even more complex models in the future. This wizard can be viewed at any time by navigating the the model editor, clicking Model in the top menu bar, and then Create → From Wizard (be aware that opening the wizard after adding any content to the model editor will clear all the content displayed and start from a new model).
Variable State Suggestion
Variables in Genius must exist in discrete states. In order to achieve this for variables representing continuous data or data of high cardinality, potential observed values must be mapped to specific states (often called "binning"). With this new feature, binning strategies for both continuous and discrete data can be used to shrink the size of probabilistic models for faster inference times and to make use of continuous data. Genius will suggest an arrangement of states for each variable that will work well for the data provided as well as a chosen strategy.
For more information about state selection and binning, see the Binning data, Building probabilistic models from scratch, and Building a POMDP model from scratch pages.
Hidden Variables
Hidden variables (also referred to as latent variables, latents, and unobserved variables - among other names) are variables which are not directly observed and hence do not get included with our data set. However, they are still important to our models and can be inferred or estimated based upon the data observations we do have. With the Darwin release of Genius, hidden variables are fully supported, with Genius able to infer probability distributions over variables for which data does not exist.
Continuous, Online Learning
Online learning allows your model to update as new data becomes available rather than requiring a full offline retraining. This online learning is also the default for POMDP models, allowing active inference agents to update the parameters of their model as new observation data becomes available. In practice, this often looks like new individual observations being provided to your agent in "real-time." This allows your agent to always have a model that accounts for the current state of the world and the most up-to-date parameters within the model.
Advanced Action Selection
POMDP action selection can now be done on multi-step plans (policy selection) and includes convenient simulation support. Currently, this feature is made available by the Python SDK and allows for users to experiment with sophisticated action-perception loops.
Self-Service Agent & Account Management
You can now manage the most critical parts of your Genius account with the VERSES Customer Portal. The Customer Portal allows you to manage agents and their licenses for quick resource allocation, add team members for collaboration, and submit tickets for new feature requests and incidents. Through the Customer Portal, you are also able to generate hosted agents, launch the model editor, and utilize the Genius SDK.
Good Housekeeping
Genius has also undergone what might typically be referred to as "general, continuous improvement" with many small pieces of work that do not warrant an expanded explanation as shown above. This includes such changes as:
Improved overall stability of the Genius software.
Enhancements to internal tooling for better validation of user-created probabilistic models.
Additions/Improvements to the Genius Python SDK for usability and to expose new Genius functionality.
General bug fixes
The Genius documentation page also been expanded to include best practices to ensure the success of the model building process. You can find out more at the frequently asked questions, tips, and tricks page.
Last updated