Abstract: One or more machine learning classifiers are trained to classify cases in one or more categories using one or more sets of labeled training data. A first distribution of scores for positive cases in the training set is determined for each category, and a second distribution of scores for negative cases in the training set is determined for each category. A third distribution of scores is generated by each classifier classifying cases in a set of target data is also determined. A proportion of cases in the target set that are positive cases for a category is estimated by fitting the first distribution and the second distribution for the category to the third distribution.

Abstract: Embodiments disclosed herein relate to providing control over a personalized category of information. In one embodiment, a personalized category of information is determined based on stored information associated with the use of an electronic device 110. In one embodiment, user feedback on the personalized category of information is received. If the user feedback comprises a rejection of the personalized category of information, the personalized category of information is discarded.

Abstract: A method and an apparatus for matching elements within sets of trajectories, locations or other attributes without revealing the entire sets. The elements are partitioned into segments. A rotating selection is made among the sets and one segment of each potentially matching element is newly disclosed from the selected set. Optionally, the sets are cryptographically hashed, using, for example, a MD5 hash or a SHA-1 hash. Optionally, the sets are represented as tries, and successively lower levels within the tries are newly disclosed from potentially matching elements as the disclosing set rotates. Optionally, the sets are encoded, using: a grid of longitude and latitude; a spatial temporal grid; a overlapping spatial grid; a temporal grid; a set of cities; a set of countries; a set of names of places; or a set of attributes. Optionally, the matching process is repeated while refining the encoding. Optionally, negotiations determine what encoding or cryptographic hash is used.

Abstract: A classification count adjustment system for adjusting a count estimate of items in a dataset D classified into a class is disclosed. The system includes a count estimate produced by a classifier of the number of items in the dataset D classified into the class. The system further comprises one or more measures of behavior of the classifier, indicating the ability of the classifier to classify items into the class. The system further comprises a processor for computing an adjusted estimate based on the count estimate by the classifier and the one or more measures of behavior.

Abstract: A related feature determination system determines a feature related to an indication of a concept. A set of positive cases and a set of negative case are selected using the indication of a concept. A classifier is induced for the concept using the set of positive cases in a manner blinded to the indication of a concept. The set of negative cases is applied to the classifier. A feature related to the indication of a concept is determined using results of applying the classifier to the set of negative cases.

Abstract: A system and method is provided for determining characteristics of sets of organized items. In an illustrative implementation, a first set of organized items as a compare-from location is identified, and a second set of organized items as a compare-to location is also identified. In an illustrative operation, a recursive intrinsic reference representing the first set of organized items is determined so that the recursive intrinsic reference is an intrinsic reference to an item containing a second intrinsic reference. A discrepancy is determined between the first set of organized items and the second set of organized items using the recursive intrinsic reference.

Abstract: A system or method for minimally predictive feature identification is disclosed. For information management, an information collection including a set of features is received. A set of prediction values indicating a degree to which a first feature within the set of features predicts other features in the set is generated. The first feature as a minimally predictive feature is identified if each of the prediction values is within a predetermined range of threshold values.

Abstract: Provided are systems, methods and techniques for machine learning. In one representative embodiment, a training set that includes training samples and corresponding assigned classification labels is obtained, and an automated classifier is trained against the training set. At least one of the training samples is selected and confirmation/re-labeling of it is requested. In response, a reply classification label is received and is used to retrain the automated classifier.

Abstract: A method for predicting parts for onsite repair which takes into account a repair history and the costs associated with mis-predictions. Parts for onsite repair of a product are predicted by determining an expected waste for one or more parts of the product. The parts having a lowest expected waste are selected and sent to the onsite repair. The expected waste indicates parts that are responsible for high support costs and highlights the mistakes being made and scores the mistakes by actual cost.

Abstract: Provided is a method for creating sketches of sets to permit comparing set members selected from a universe. The method includes selecting a size for the sketch and providing a randomizer. In addition, a set is provided as a subset of the universe. The set is processed with the randomizer to provide a new vector. The new vector is normalized to provide a value. Dividing each element of the new vector by the determined value results in the sketch.

Abstract: A method is provided for the identification of managed systems that are exceptional compared to comparable managed systems. The method involves selecting a set of managed systems wherein each managed system has a number of system configuration attributes and selecting a set of parameterizations relating to the managed systems. Patterns are determined for the parameterizations within the set of managed systems using a supervised machine learning algorithm. The managed systems are compared to the patterns and any managed systems that deviate from the patterns are isolated as exceptional managed systems. Isolated systems may be targeted for remedial action such as the allocation of support resources or system management resources.

Abstract: A count of cases in a target set classified in a class is computed using a classifier. In the method, one or more measures of behavior of the classifier are determined for a plurality of classification thresholds and at least one classification threshold for the classifier is selected based on the one or more measures of behavior. In addition, a score for a plurality of the cases in the target set using the classifier is computed and the count of the cases is computed based on the scores, the selected at least one classification threshold, and the one or more measures of behavior.

Abstract: The present invention is a system or method for assisting in the maintenance and servicing of computers. The system, for example, comprises a cluster database containing information relating to one or more computer clusters, the information relating to each cluster comprising one or more cluster descriptions and one or more representations of configuration information values that characterize typical computer members of the cluster, and also a tracker database containing configuration information values gathered from and relating to one or more computers.

Abstract: Provided are systems, methods and techniques for machine learning. In one representative embodiment, a training set that includes training samples and corresponding assigned classification labels is obtained, and an automated classifier is trained against the training set. At least one of the training samples is selected and confirmation/re-labeling of it is requested. In response, a reply classification label is received and is used to retrain the automated classifier.

Abstract: Provided is a method for creating sketches of sets to permit comparing set members selected from a universe. The method includes selecting a size for the sketch and providing a randomizer. In addition, a set is provided as a subset of the universe. The set is processed with the randomizer to provide a new vector. The new vector is normalized to provide a value. Dividing each element of the new vector by the determined value results in the sketch.

Abstract: A system and method for customer-side market segmentation and categorization. This segmentation is done without disclosing sensitive private customer information to the business. A customer downloads a categorization module to a portable device (PDA, wireless cellular phone, etc.) or personal computer. A business defines a decision procedure corresponding to a set of defined customer categories. The business sends their rule set to the customer's device, which uses the rules and a set of stored customer-specific historical and demographic information to determine into which of the business-specific customer categories the customer falls. The categorization module may use any of a variety of methods, such as decision trees, neural networks, Bayesian belief networks, k-nearest neighbor, genetic algorithms, or rule sets. The customer category is sent to the business without other personal data for the business to prepare appropriate promotional material or initiate specific actions.

Abstract: The present invention relates generally to the classification of items into categories, and more generally, to the automatic selection of different classifiers at different places within a hierarchy of categories. An exemplary hierarchical categorization method uses a hybrid of different classification technologies, with training-data based machine-learning classifiers preferably being used in those portions of the hierarchy above a dynamically defined boundary in which adequate training data is available, and with a-priori classification rules not requiring any such training-data being used below that boundary, thereby providing a novel hybrid categorization technology that is capable of leveraging the strengths of its components.

Abstract: A system and method for category organization is disclosed. The method discloses: receiving an information collection including a set of strings; identifying positively predictive pairs of strings; identifying negatively predictive pairs of strings; joining positively predictive pairs of strings into common categories; splitting negatively predictive pairs of strings into different categories; and organizing the categories using the negatively predictive pairs of strings. The system discloses various elements, means and instructions for performing the method.

Abstract: A system and method for category discovery is disclosed. The method discloses: receiving an information collection including a set of strings; identifying positively predictive pairs of strings; identifying negatively predictive pairs of strings; joining positively predictive pairs of strings into a category; and splitting negatively predictive pairs of strings into different categories. The system discloses various elements, means and instructions for performing the method.

Abstract: Data associated with at least one characteristic associated with the viability of a product is monitored. The data associated with the at least one characteristic is analyzed, and based on the analyzing, at least one future viability state condition of the product is predicted. At least one indicator is displayed related to the at least one future viability state condition.