Focus and Bias: Will It Blend?

One direct application of explainable AI feature attribution methods is to be used for detecting unwanted biases. To do so, domain experts typically have review explained inputs, checking the presence biases learnt by model. However, huge amount samples must makes this task more challenging as size dataset grows. In an ideal case, should provided only with a small number selected containing potential The recently published Focus score seems promising tool selection work, we conduct first study in direction, analyzing behavior when applied biased First, verified that indeed sensitive induced bias. This assessed forcing spurious correlation, training model using cats-indoor and dogs-outdoor. We empirically prove distinguish contexts (outdoor vs indoor) instead cat dog classes, so ensuring Afterwards, apply on showing how decreases input contains aforementioned analysis sheds light model, highlighting its strengths use bias detection.