Building a better AI data engine

On Monday, July 12th, Labelbox CEO and Co-founder Manu Sharma presented a keynote session at VentureBeat Transform. The day's events were dedicated to exploring the technologies that enable automation for AI and ML. Read on for an overview of the Labelbox session, Building a better AI data engine.

Model assisted labeling (MAL)

This feature takes a "fully aware" approach to automation, enabling Labelbox customers to import their model's output as pre-labeled training data. This allows ML teams to review and correct these datasets, as well as use them to retrain the model to improve accuracy. This technique ensures that each iteration requires less human effort, helping ML teams reduce costs and increasing efficiency.

The VentureBeat Transform session will cover use cases and results from customers who have used the MAL feature with success.

Model diagnostics

Labelbox's newest set of platform features tackles one of the biggest stumbling blocks to creating high quality data. With it, ML teams use the Labelbox platform to:

• Pinpoint errors in model output by comparing it visually with the ground truth
• Identify patterns in model behavior over iterations
• Correct errors

With model diagnostics, ML teams can be sure that their training datasets are optimized to drive model performance improvements. Model diagnostics is currently in closed beta; you can learn more and sign up for the beta waitlist here.

Learn more about MAL, model diagnostics, and our philosophy behind iteration and data quality at the VentureBeat Transform session presented by Labelbox.