Virtual

Americas

Workshops

What’s Hiding Behind Your F1-Score? How to Uncover Hidden Model Failures

Oct 9, 2025

10 AM Pacific

Online. Register for the Zoom!

About this event

Learn how to uncover hidden weaknesses in AI models before they reach production with FiftyOne’s data-centric workflows. This hands-on session features real-world demos across industries and practical strategies for debugging failures, improving label quality, and strengthening model performance.

Host

Seemingly “accurate” AI models can stumble when they meet real-world data. In this hands-on session, we’ll show you how FiftyOne’s data-centric workflows expose hidden weaknesses before they reach production. Through a live demo using real-world use cases across Robotics, Manufacturing, Autonomous Vehicles, and more,, you'll learn practical strategies to quickly uncover data gaps, debug common failure modes, and harden your models.

What you’ll learn:

Why metrics can be misleading: Discover how the most commonly used metrics for model success can mask critical issues, and how scenario-based, customizable model evaluation workflows can illuminate hidden failures.
Edge-case detection: Use embeddings and custom scenarios to reveal missing conditions and underrepresented classes.
Better label QA: Easily identify noisy, missing, and inconsistent labels with FiftyOne’s Verified Auto Labeling
Capturing the long tail: Balance rare classes with targeted scenario analysis, augmentation, and synthetic data to improve model generalization and reduce bias
Continuous model improvement: Track data drift and model confidence to ensure training data continues to represent real-world conditions

Who should attend:

Machine learning engineers, QA engineers, project managers, and anyone interested in shipping better AI models quickly, with visibility and control.