We’re a few months past SAS Innovate and the clarity that comes from not being in the middle of it all gives the chance to see how good we had it. Luckily, we also now have recordings from over 100 of the conference sessions as well. I know, I know … this is yet another conference recap—but this one’s for those of your who are interested in models and model deployment, management, and governance.
Session highlights
Responsible AI: Building ModelOps Processes with SAS Model Manager
In this session, Sophia Rowland dives into the responsible AI features built into SAS Model Manager. The focus is all about AI governance and how SAS helps ensure responsible AI practices throughout the model life cycle.
Sophia’s demo covered some standout features:
- Model Cards: Like a nutrition label for models—includes training data, performance metrics, fairness assessments, and usage guidelines.
- Performance monitoring: Keeps an eye on things like input/output distribution changes, ROC curves, and misclassification rates.
- Fairness & bias monitoring: New as of March—assesses model behavior across different demographic groups.
- Model publish locking: Prevents unauthorized or premature model deployment
- The (now open source!) Trustworthy AI Lifecycle Workflow, based on the NIST AI Risk Management Framework, available both in SAS Viya and via GitHub as open source.
Check out the recording
Data Ethics and AI Governance
Reggie Townsend, the VP of Data Ethics Practice, and Allie DeLonay, a Senior Data Scientist, dive into the big topics around AI responsibility, governance, and oversight. They share why these issues are crucial for modern businesses and even give some real-world examples of challenges in AI governance and how SAS’s internal AI Oversight Committee steps in to tackle them.
Allie also gives a demo of the AI Governance Navigator, an upcoming tool designed to make it easier to keep track of, evaluate, and manage your AI assets. Fun fact: this tool is set to be available for private preview later this year.
Watch the recording
Using SAS Model Manager to Deploy Models Trained in SAS Viya Workbench
Sophia’s back—showing how to deploy SAS Viya Workbench-trained models in SAS Model Manager. Here, she covers both SAS and Python workflows and her session emphasizes integration, model registration, and post-deployment capabilities.
Watch the demo to see the model deployment workflow, from SAS Viya Workbench and Python to SAS Model Manager.
Some of the big takeaways here are that SAS model Manager:
- Support multi-language model integration.
- Enables governed ModelOps/MLOps processes.
- Offers robust monitoring and fairness tools to ensure your model’s reliability and ethical use.
Here’s that recording
Other things you might have missed
SAS Models
By now, you’ve surely seen all about SAS Models: ready-made, lightweight, industry-specific models. First time hearing about this? Here’s a blog to get you up to speed.
At SAS Innovate, we had a SAS Models booth where attendees could learn about the models we’ve already released, as well as some of the upcoming announcements. Of course, everyone’s talking about agentic AI, but the standout model, in my opinion, was the Data Mapper Agent. Not to dumb it down too much, but this agent uses an LLM to automatically map columns of your existing data to columns required by AI models.
Really exciting stuff.
There’s a lot more out there from SAS Innovate. You can check out all the other recordings, and more, on our YouTube channel. or on the SAS Communities.
