MAIA can also be applied to surface model-level biases. In a preliminary demonstration, we investigate biases in the outputs of an image classifier (ResNet-152) trained on a supervised ImageNet classification task. MAIA is easily adapted to this task by instrumenting an output logit corresponding to a particular image class as the system to be interpreted, and instructing MAIA to describe the actual distribution of images that receive high class scores, and whether that correctly matches the class label (see ppaer Appendix for full MAIA instructions).
+MAIA can also be applied to surface model-level biases. In a preliminary demonstration, we investigate biases in the outputs of an image classifier (ResNet-152) trained on a supervised ImageNet classification task. MAIA is easily adapted to this task by instrumenting an output logit corresponding to a particular image class as the system to be interpreted, and instructing MAIA to determine whether the actual distribution of images that receive high class scores matches the class label (see papaer Appendix for full MAIA instructions).
