> you can find out if someone is in the training set by simply measuring the error rate on that person.
I don't think it's that easy to detect if a sample exists in the training set. Error rate are usually informative when measured across multiple predictions. With a single prediction you're most likely looking at the estimated probability / confidence (with Logistic Regression for example). In that case models might assign high probability to unseen inputs as well.
I don't think it's that easy to detect if a sample exists in the training set. Error rate are usually informative when measured across multiple predictions. With a single prediction you're most likely looking at the estimated probability / confidence (with Logistic Regression for example). In that case models might assign high probability to unseen inputs as well.