Technique for identifying features

A non-transitory computer-readable storage medium for use in conjunction with an electronic device, the computer-readable storage medium storing program instructions that, when executed by the electronic device, causes the electronic device to perform one or more operations comprising:

• calculating combinations of feature vectors and noise vectors, wherein a given combination in the combinations corresponds to a given feature vector in the feature vectors and a given noise vector in the noise vectors, wherein a number of the feature vectors exceeds one hundred and a number of the combinations exceeds ten thousand;
• determining statistical associations between types of events and the combinations, wherein a given statistical association in the statistical associations corresponds to the types of events and the given combination;
• computing a noise threshold associated with a group of combinations in the combinations, wherein the noise threshold corresponds to a first aggregate property of the group of combinations, wherein the first aggregate property comprises numbers of occurrences of the feature vectors in the group of combinations, and wherein the group of combinations have statistical associations equal to or greater than the noise threshold;
• selecting a subset of the feature vectors based at least in part on the first aggregate property of the group of combinations;
• and
• training a supervised machine-learning model using the types of events and a training dataset that is based at least in part on the selected subset of the feature vectors, wherein training the supervised machine-learning model comprises determining parameters in the supervised machine-learning model and identifying feature vectors in the subset of feature vectors to use in the supervised machine-learning model, and wherein the supervised machine-learning model predicts the types of events based at least in part on the identified feature vectors and the parameters.

A method for selecting a subset of feature vectors, comprising:

• by an electronic device:
• calculating combinations of feature vectors and noise vectors, wherein a given combination in the combinations corresponds to a given feature vector in the feature vectors and a given noise vector in the noise vectors, wherein a number of the feature vectors exceeds one hundred and a number of the combinations exceeds ten thousand;
• determining statistical associations between types of events and the combinations, wherein a given statistical association in the statistical associations corresponds to the types of events and the given combination;
• computing a noise threshold associated with a group of combinations in the combinations, wherein the noise threshold corresponds to a first aggregate property of the group of combinations, wherein the first aggregate property comprises numbers of occurrences of the feature vectors in the group of combinations, and wherein the group of combinations have statistical associations equal to or greater than the noise threshold;
• selecting a subset of the feature vectors based at least in part on the first aggregate property of the group of combinations;
• and
• training a supervised machine-learning model using the types of events and a training dataset that is based at least in part on the selected subset of the feature vectors, wherein training the supervised machine-learning model comprises determining parameters in the supervised machine-learning model and identifying feature vectors in the subset of feature vectors to use in the supervised machine-learning model, and wherein the supervised machine-learning model predicts the types of events based at least in part on the identified feature vectors and the parameters.