Nonconformity Score
A metric in conformal prediction that measures the discrepancy between a model's prediction and the true label. For regression tasks, absolute error is commonly used, while for classification tasks, one minus the predicted probability of the true class is typically employed.
Timeline (last 90 days)
Conformal prediction relies on two key concepts: nonconformity scores and a calibration set, where the calibration set is data with true labels that is independent of the training set.
A common nonconformity score definition in classification tasks is 1 minus the probability value corresponding to the true class in the softmax output, where a higher score indicates a more anomalous sample.
All Facts (2)
Conformal prediction relies on two key concepts: nonconformity scores and a calibration set, where the calibration set is data with true labels that is independent of the training set.
50%UnverifiedA common nonconformity score definition in classification tasks is 1 minus the probability value corresponding to the true class in the softmax output, where a higher score indicates a more anomalous sample.
50%