Vertex Explainable AI is a tool designed to enhance the understanding and trust in machine learning models by providing insights into their decision-making processes. It offers two primary methods for explaining model behavior: feature-based explanations, which highlight how each feature contributes to a prediction, and example-based explanations, which identify similar instances from the training data to explain predictions.

Key Features

Feature-based Explanations

• Provide attribution scores for each feature in a model.
• Supported for various model types and frameworks (TensorFlow, scikit, AutoML).
• Methods include integrated gradients, sampled Shapley, and XRAI.

Example-based Explanations

• Utilize nearest neighbor search to find similar instances from the training dataset.
• Helpful for identifying model mistakes, interpreting novel data, detecting anomalies, and active learning.
• Currently supports TensorFlow models.

Benefits

• Model Improvement: Helps identify issues in models and data, enabling targeted improvements.
• Prediction Confidence: Enhances trust in model predictions by providing insights into decision-making processes.
• Anomaly Detection: Facilitates the identification of outliers that may not be well-represented in the training data.