Abstract: In various embodiments, a training application generates a preference prediction model based on an interaction matrix and a closed-form solution for minimizing a Lagrangian. The interaction matrix reflects interactions between users and items, and the Lagrangian is formed based on a constrained optimization problem associated with the interaction matrix. A service application generates a first application interface that is to be presented to the user. The service application computes predicted score(s) using the preference prediction model, where each predicted score predicts a preference of the user for a different item. The service application then determines a first item from the items to present to the user via an interface element included in the application interface. Subsequently, the service application causes a representation of the first item to be displayed via the interface element included in the application interface.

Abstract: In various embodiments, a training application generates a preference prediction model based on an interaction matrix and a closed-form solution for minimizing a Lagrangian. The interaction matrix reflects interactions between users and items, and the Lagrangian is formed based on a constrained optimization problem associated with the interaction matrix. A service application generates a first application interface that is to be presented to the user. The service application computes predicted score(s) using the preference prediction model, where each predicted score predicts a preference of the user for a different item. The service application then determines a first item from the items to present to the user via an interface element included in the application interface. Subsequently, the service application causes a representation of the first item to be displayed via the interface element included in the application interface.

Abstract: Techniques are described for promoting original media titles. Given metadata tags associated with the original title and other media titles, a tag data matrix is generated and factored into two matrices, one of which includes vectors representing the media titles in a first latent space. Similarity scores are computed between a vector representing the original title and each of the other media title vectors to determine a set of media titles most similar to the original title. Then, a play data matrix is factorized, and an average of vectors representing the most similar titles in a second latent space is taken to be a vector representation of the original title in the second latent space. This representation is compared with representations of users in the second latent space to generate similarity scores, and the original title is then promoted to users associated with the highest similarity scores.

Abstract: In one embodiment of the present invention, a training engine teaches a matrix factorization model to rank items for users based on implicit feedback data and a rank loss function. In operation, the training engine approximates a distribution of scores to corresponding ranks as an approximately Gaussian distribution. Based on this distribution, the training engine selects an activation function that smoothly maps between scores and ranks. To train the matrix factorization model, the training engine directly optimizes the rank loss function based on the activation function and implicit feedback data. By contrast, conventional training engines that optimize approximations of the rank loss function are typically less efficient and produce less accurate ranking models.

Abstract: In one embodiment of the present invention, a training engine teaches a matrix factorization model to rank items for users based on implicit feedback data and a rank loss function. In operation, the training engine approximates a distribution of scores to corresponding ranks as an approximately Gaussian distribution. Based on this distribution, the training engine selects an activation function that smoothly maps between scores and ranks. To train the matrix factorization model, the training engine directly optimizes the rank loss function based on the activation function and implicit feedback data. By contrast, conventional training engines that optimize approximations of the rank loss function are typically less efficient and produce less accurate ranking models.

Abstract: A processing device of an information processing system is operative to obtain observed feedback data, to construct a model that accounts for both the observed feedback data and additional feedback data that is missing from the observed feedback data, to optimize one or more parameters of the model using a training objective function, and to generate a list of recommended items for a given user based on the optimized model. In illustrative embodiments, the missing feedback data comprises data that is missing not at random (MNAR), and the model comprises a matrix factorization model. The processing device may implement a recommender system comprising a training module coupled to a recommendation module.

Abstract: Techniques are described for promoting original media titles. Given metadata tags associated with the original title and other media titles, a tag data matrix is generated and factored into two matrices, one of which includes vectors representing the media titles in a first latent space. Similarity scores are computed between a vector representing the original title and each of the other media title vectors to determine a set of media titles most similar to the original title. Then, a play data matrix is factorized, and an average of vectors representing the most similar titles in a second latent space is taken to be a vector representation of the original title in the second latent space. This representation is compared with representations of users in the second latent space to generate similarity scores, and the original title is then promoted to users associated with the highest similarity scores.

Abstract: An exemplary method of establishing a decision tree includes determining an effectiveness indicator for each of a plurality of input features. The effectiveness indicators each correspond to a split on the corresponding input feature. One of the input features is selected as a split variable for the split. The selection is made using a weighted random selection that is weighted according to the determined effectiveness indicators.

Abstract: Medical related quality of care information is extracted and edited for reporting. Patient records are mined. The mining may include mining unstructured data to create structured information. Measures are derived automatically from the structured information. A user may then edit the measures, data points used to derive the measures, or other quality metric based on expert review. The editing may allow for a better quality report. Tools may be provided to configure reports, allowing generation of new or different reports.

Abstract: The invention concerns a method and a server for routing between devices of a computer based social network having a plurality of users, wherein upon receipt of a first message from a device (110) associated with a user (u1), a second message is sent to another device (104, 110), wherein the other device (104, 110) is selected from a plurality of predetermined devices depending on the result of an evaluation of at least 1 trust value (S(c,u1;uM)) associated with the user (u1), a category (c) of content of the computer based social network and another user (uM) of the computer based social network.

Abstract: Embodiments are directed to mobile localization, and more specifically, but not exclusively, to tracking mobile devices. Embodiments include methods that consider probability kernels with distance-like metrics between distributions. Also described are probabilistic kernels that can be used for a regression of location, which can achieve up to about inn accuracy in an office environment.

Abstract: Embodiments are directed to mobile localization, and more specifically, but not exclusively, to tracking mobile devices. Embodiments include methods that consider probability kernels with distance-like metrics between distributions. Also described are probabilistic kernels that can be used for a regression of location, which can achieve up to about inn accuracy in an office environment.

Abstract: A processing device of an information processing system is operative to obtain observed feedback data, to construct a model that accounts for both the observed feedback data and additional feedback data that is missing from the observed feedback data, to optimize one or more parameters of the model using a training objective function, and to generate a list of recommended items for a given user based on the optimized model. In illustrative embodiments, the missing feedback data comprises data that is missing not at random (MNAR), and the model comprises a matrix factorization model. The processing device may implement a recommender system comprising a training module coupled to a recommendation module.

Abstract: A method for predicting survival rates of medical patients includes providing a set D of survival data for a plurality of medical patients, providing a regression model having an associated parameter vector ?, providing an example x0 of a medical patient whose survival probability is to be classified, calculating a parameter vector {circumflex over (?)} that maximizes a log-likelihood function of ? over the set of survival data, l(?|D), wherein the log likelihood l(?|D) is a strictly concave function of ? and is a function of the scalar x?, calculating a weight w0 for example x0, calculating an updated parameter vector ?* that maximizes a function l(?|D?{(y0,x0,w0)}), wherein data points (y0,x0,w0) augment set D, calculating a fair log likelihood ratio ?f from {circumflex over (?)} and ?* using ?f=?(?*|x0)+sign(?({circumflex over (?)}|x0)){l({circumflex over (?)}|D)?l(?*|D)}, and mapping the fair log likelihood ratio ?f to a fair price y0f, wherein said fair price is a probability that class label y0 for exam

Abstract: The present invention provides methods and compositions for predicting patient responses to cancer treatment using a proliferation gene signature. These methods can comprise measuring in a biological sample from a patient the levels of gene expression of a group of the genes designated herein. The present invention also provides for microarrays that can detect expression from a group of genes.

Abstract: According to an aspect of the invention, a method for training a classifier for classifying candidate regions in computer aided diagnosis of digital medical images includes providing a training set of annotated images, each image including one or more candidate regions that have been identified as suspicious, deriving a set of descriptive feature vectors, where each candidate region is associated with a feature vector. A subset of the features are conditionally dependent, and the remaining features are conditionally independent. The conditionally independent features are used to train a naïve Bayes classifier that classifies the candidate regions as lesion or non-lesion. A joint probability distribution that models the conditionally dependent features, and a prior-odds probability ratio of a candidate region being associated with a lesion are determined from the training images.

Abstract: A medical concept is learned about or inferred from a medical transcript. A probabilistic model is trained from medical transcripts. For example, the problem is treated as a graphical model. Discrimitive or generative learning is used to train the probabilistic model. A mutual information criterion can be employed to identify a discrete set of words or phrases to be used in the probabilistic model. The model is based on the types of medical transcripts, focusing on this source of data to output the most probable state of a patient in the medical field or domain. The learned model may be used to infer a state of a medical concept for a patient.

Abstract: Medical ontology information is used for mining and/or probabilistic modeling. A domain knowledge base may be automatically or semi-automatically created by a processor from a medical ontology. The domain knowledge base, such as a list of disease associated terms, is used to mine for corresponding information from a medical record. The relationship of different terms with respect to a disease may be used to train a probabilistic model. Probabilities of a disease or chance of indicating the disease are determined based on the terms from a medical ontology. This probabilistic reasoning is learned with a machine from ontology information and a training data set.

Abstract: A predictor of medical treatment outcome is developed and applied. A prognosis model is developed from literature. The model is determined by reverse engineering the literature reported quantities. A relationship of a given variable to a treatment outcome is derived from the literature. A processor may then use individual patient values for one or more variables to predict outcome. The accuracy may be increased by including a data driven model in combination with the literature driven model.

Abstract: Medical ontology information is used for mining and/or probabilistic modeling. A domain knowledge base may be automatically or semi-automatically created by a processor from a medical ontology. The domain knowledge base, such as a list of disease associated terms, is used to mine for corresponding information from a medical record. The relationship of different terms with respect to a disease may be used to train a probabilistic model. Probabilities of a disease or chance of indicating the disease are determined based on the terms from a medical ontology. This probabilistic reasoning is learned with a machine from ontology information and a training data set.