LEARNDEEPLEARNINGWITHME.WORDPRESS.COM

With "Deep Learning" itself being a buzzword over recent years, many researchers now try to pinpoint to the answer to: "Can we trust a model simply based on the metrics of interest?"... "Why Should I Trust You?" is one of the pioneer research papers in the arena of Interpretable Deep Learning. Their work is solely based on the hypothesis that the trust developed on a model by the users impacts the acceptance and usage of the model... How often we give any thoughts about what really happens inside an ML model for it to make a classification or prediction? Currently, all models are evaluated using metrics such as Accuracy/Recall/Prediction on the validation data available and the relationship between the prediction and the input features still remains inside a black-box. The increase in stakes involved in blindly trusting a prediction/model demands evaluating a model based on trust. To develop trust we try to explain human users how the model arrived to this prediction by making the..