Abstract: Machine learning adversarial campaign mitigation on a computing device. The method may include deploying an original machine learning model in a model environment associated with a client device; deploying a classification monitor in the model environment to monitor classification decision outputs in the machine learning model; detecting, by the classification monitor, a campaign of adversarial classification decision outputs in the machine learning model; applying a transformation function to the machine learning model in the model environment to transform the adversarial classification decision outputs to thwart the campaign of adversarial classification decision outputs; determining a malicious attack on the client device based in part on detecting the campaign of adversarial classification decision outputs; and implementing a security action to protect the computing device against the malicious attack.

Abstract: Identifying and protecting against a computer security threat while preserving privacy of individual client devices using differential privacy for text documents. In some embodiments, a method may include receiving, at the remote server device, text documents from one or more local client devices, generating, at the remote server device, a differential privacy document vector for each of the text documents, identifying, at the remote server device, a computer security threat to a first one of the one or more local client devices using the differential privacy document vectors, and, in response to identifying the computer security threat, protecting against the computer security threat by directing performance, at the first local client device or the remote server device, of a remedial action to protect the first local client device from the computer security threat.

Abstract: Identifying and protecting against a computer security threat while preserving privacy of individual client devices using differential privacy machine learning for streaming data. In some embodiments, a method may include receiving first actual data values streamed from one or more first local client devices, generating first perturbed data values by adding noise to the first actual data values using a differential privacy mechanism, storing the first perturbed data values, training a machine learning classifier using the first perturbed data values, receiving a second actual data value streamed from a second local client device, generating a second perturbed data value by adding noise to the second actual data value, storing the second perturbed data value, identifying a computer security threat to the second local client device using the second actual data value as input to the trained machine learning classifier, and protecting against the computer security threat.

Abstract: Verifying that influence of a user data point has been removed from a machine learning classifier. In some embodiments, a method may include training a machine learning classifier using a training set of data points that includes a user data point, calculating a first loss of the machine learning classifier, updating the machine learning classifier by updating parameters of the machine learning classifier to remove influence of the user data point, calculating a second loss of the machine learning classifier, calculating an expected difference in loss of the machine learning classifier, and verifying that the influence of the user data point has been removed from the machine learning classifier by determining that the difference between the first loss and the second loss is within a threshold of the expected difference in loss.

Abstract: Verifying that influence of a user data point has been removed from a machine learning classifier. In some embodiments, a method may include training a machine learning classifier using a training set of data points that includes a user data point, calculating a first loss of the machine learning classifier, updating the machine learning classifier by updating parameters of the machine learning classifier to remove influence of the user data point, calculating a second loss of the machine learning classifier, calculating an expected difference in loss of the machine learning classifier, and verifying that the influence of the user data point has been removed from the machine learning classifier by determining that the difference between the first loss and the second loss is within a threshold of the expected difference in loss.