Maximizing Model Accuracy: Strategies from ML Experts

High model accuracy is hard to establish — and sometimes harder to maintain. Once you’ve trained a model, your work is just beginning. In fact, 91% of ML models degrade over time once they’re in the real world. So for many ML teams, it’s just not feasible to meet optimal accuracy targets. How do you improve accuracy without straining your resources?

Join us for a panel talk on Tuesday, Dec. 17 with Tecton CTO and co-founder Kevin Stumpf, and founder of Decoding ML Paul Iusztin, where they will discuss the real-world challenges of model accuracy in production ML. They’ll also explore architectural patterns and share best practices for achieving higher accuracy.

Drawing from their combined experience building production systems across Uber, Continental, Metaphysic.AI, CoreAI and Everseen, this technical talk will cover common questions like:

• What are the “silent killers” of accuracy that make good models go bad?
• What goes wrong when you go from training to serving? Why does seemingly accurate training data underperform in production?
• How do you balance accuracy goals with resource constraints, such as infrastructure costs and available talent?
• How does feature production make an impact on model accuracy? 

Don't miss this opportunity to learn from real-world experiences and get practical advice for your own accuracy initiatives. Whether you're scaling your ML use cases or troubleshooting existing models, you'll walk away with strategies for improving model accuracy in production, including more efficient feature engineering.