3 ms·
One idea: try using adversarial training (i.e. train the classifier on examples with bounded perturbations which maximize error).
by alexmlamb 7y ago
One idea: try using adversarial training (i.e. train the classifier on examples with bounded perturbations which maximize error).