Explaining Classification – counterfactuals and gradients

What is the goal of CROCO?

To generate robust counterfactual explanations that remain valid under small changes.

What is the goal of DeepLIFT?

To provide accurate feature attributions by decomposing a model’s prediction into contributions from each input.

What is the goal of Stable Neighbor Search (SNS)?

To find consistent counterfactual explanations that remain valid across model retraining or slight changes.

What problem does CROCO solve?

Counterfactual explanations often fail when small changes (perturbations) occur in input features. CROCO ensures robustness.