QuestionQ53

Serving and scaling models

You work for a social media company and need to determine whether posted images contain cars. Every training example belongs to exactly one class. You trained an object-detection neural network and deployed its model version to AI Platform Prediction for evaluation. Before deploying it, you created an evaluation job and attached it to the AI Platform Prediction model version. You find that precision is below what the business requirements permit. How should you change the final-layer softmax threshold to raise precision?

Explanation

Increasing a confidence threshold returns only higher-scoring predictions. This typically improves precision by filtering out false positives, while reducing recall because some true positives no longer meet the threshold.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!