Abstract: Techniques for predicting a number of links an email campaign recipient will open are described. Elements in a dataset related to an email campaign are modeled using a tree structure, where nodes of the tree represent features of each element. A mean squared error is computed of an outcome for each of the elements to determine a weight for each respective tree. The weights are then regularized by applying a penalty, such as an elastic net penalty, to each of the weights. Then, the weights are applied to each of the trees. A weighted average of all of the outcomes of the trees is calculated, where the weighted average represents a prediction of an outcome resulting from a set of feature values. The feature values correspond to the nodes of each of the trees.

Abstract: Certain embodiments involve a model for predicting user behavior. For example, a system accesses user behavior data indicating various users' behaviors during intervals over various periods of time and target behavior data indicating a particular user behavior. The system associates each user with a label that indicates whether a user performed a particular action during or after a time period based on the target behavior data. The system uses the user behavior data to train various deep Restricted Boltzmann Machines (“RBM”) to generate representations of each user over each period of time that indicate the user behavior over the time period. The system generates a predictive model by connecting the RBMs into a deep recurrent neural network and uses the target behavior data associated with each user, along with the representations of each user, as input data to train the deep recurrent neural network to predict user behavior.

Abstract: A method for updating the resistance of a controllable resistance element includes determining an amount of resistance change for the controllable resistive element. A charge difference for a battery is determined corresponding to the resistance change for the controllable resistive element. The battery is charged or discharged to effect the resistance change in the controllable resistive element.

Abstract: A device may receive training spectral data associated with a manufacturing process that transitions from an unsteady state to a steady state. The device may generate, based on the training spectral data, a plurality of iterations of a support vector machine (SVM) classification model. The device may determine, based on the plurality of iterations of the SVM classification model, a plurality of predicted transition times associated with the manufacturing process. A predicted transition time, of the plurality of predicted transition times, may identify a time, during the manufacturing process, that a corresponding iteration of the SVM classification model predicts that the manufacturing process transitioned from the unsteady state to the steady state. The device may generate, based on the plurality of predicted transition times, a final SVM classification model associated with determining whether the manufacturing process has reached the steady state.

Abstract: Provided is a method of machine learning for a convolutional neural network (CNN). The method includes: receiving input target data; determining whether to initiate incremental learning on the basis of a difference between a statistical characteristic of the target data with respect to the CNN and a statistical characteristic of previously used training data with respect to the CNN; determining a set of kernels with a high degree of mutual similarity in each convolution layer included in the CNN when the incremental learning is determined to be initiated; and updating a weight between nodes to which kernels included in the set of kernels with a high degree of mutual similarity are applied.

Abstract: In some examples, structured and unstructured data is evaluated using one or more predictive models to determine whether a dependent user is at risk for a certain condition. In other examples, structured and unstructured data is evaluated using one or more predictive models to determine a contact plan for contacting dependent users regarding follow-up appointments related to release of the dependent user.

Abstract: The technology disclosed relates to an end-to-end neural network for question answering, referred to herein as “dynamic coattention network (DCN)”. Roughly described, the DCN includes an encoder neural network and a coattentive encoder that capture the interactions between a question and a document in a so-called “coattention encoding”. The DCN also includes a decoder neural network and highway maxout networks that process the coattention encoding to estimate start and end positions of a phrase in the document that responds to the question.

Abstract: A method, system, and computer program product for social collaboration in probabilistic prediction are provided in the illustrative embodiments. A set of predictions is sent to a user device. A prediction in the set of predictions is a probability of an outcome of an event. The probability is computed using a prediction model trained with training data corresponding to the event. An input is received from the user device. The input comprises a new prediction made at the user device using a new prediction model executing on the user device. A difference is determined between the prediction and the new prediction. The prediction model is revised to produce a revised prediction. A revised difference between the revised prediction and the new prediction is smaller than the difference.

Abstract: A method, system and computer program product for enhancing privacy of event data. Event sensor data (e.g., body temperature, heart rate data) is received and analyzed by a subscriber to form a probability of assigning a user identity to the received event sensor data. The user of the event sensor data is then assigned with a temporary membership to a cohort (related group of users that share common characteristic(s) or experience(s)) to hide the identity of the user in response to the probability of assigning the user identity to the received event sensor data exceeding a threshold. Actions may then be performed based on the temporary membership to the cohort in order to ensure that the probability of assigning a user identity to the received event sensor data does not exceed the threshold. In this manner, privacy of the user's sensitive data is enhanced.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for efficiently processing dynamic length tensors of a machine learning model represented by a computational graph. A program is received that specifies a dynamic, iterative computation that can be performed on input data for processing by a machine learning model. A directed computational graph representing the machine learning model is generated that specifies the dynamic, iterative computation as one or more operations using a tensor array object. Input is received for processing by the machine learning model and the directed computational graph representation of the machine learning model is executed with the received input to obtain output.

Abstract: An output time-series of a cell of a neural network is captured. A subset of a set of data points of the output time-series is consolidated into a singular data point. The singular data point is fitted in a data representation to form a quantified aggregated data point. The quantified aggregated data point is included in an intermediate time-series. Using the intermediate time-series as an input at an intermediate layer of the neural network, an anonymized output time-series is produced from the neural network.

Abstract: The present disclosure relates to a system, method, and computer program for predicting migraines using machine learning. More specifically, the system predicts the likelihood that a user will have a migraine within a certain period of time (e.g., a month) based on migraine causes/triggers X1 . . . n identified from a migraine pattern built using deep neural networks. The system collects data from the user with respect to his/her profile, his/her migraine incidents (having experienced a migraine or not) and his/her daily habits. Based on these data, it creates the user's migraine pattern and combines this pattern with the migraine patterns from similar users (when necessary). As a result, it is able to predict the probability of getting a migraine within a period of time.

Abstract: Feature engineering can be performed on time series data making the data easy to manipulate and accessible to business users for analysis according to existing best practices. A computer system can, after receiving time series data related to a device, contextualize the time series data based on business data related to the device from, for example, an enterprise resource planning database. The contextualized data can be windowed by a selected feature based on execution data related to the device from, for example, a manufacturing execution system database. The windowed data can be transformed into summary data using a time series transformation. The summary data can be easily manipulated by, for example, generating genetic maps of the segmented and transformed data for clustering or searching for anomalies and patterns in response to user requests or automatically.

Abstract: This disclosure relates to systems and methods for performing an autonomous procedure for monitoring and diagnostics of a machine using electrical signature analysis. In one embodiment of the disclosure, a method includes providing electrical data of an electrical rotating machine associated with at least one fault frequency. While in a learning mode, the method includes converting the electrical data from a time domain to a frequency domain to obtain baseline data. While in an operational mode, the method includes converting the electrical data from the time domain to the frequency domain to obtain monitoring data. The method further includes determining, based at least on the monitoring data, a ratio value at the fault frequency, determining a rate of change of the ratio value at the fault frequency, and, optionally, issuing, based on the rate of change, an alarm concerning at least one event of the electrical rotating machine.

Abstract: A method, system and computer program product for generating a solution to an optimization problem. A received structured set of data is analyzed with the prescriptive domains to identify one or more prescriptive domains that match the received structure set of data in data structure and/or semantic terms. A user selection of one of the presented possible prescriptive intentions from the intention templates in the identified one or more prescriptive domains that match the received structure set of data in data structure and/or semantic terms is received. A prescriptive model is then generated from the prescriptive domain containing the selected prescriptive intention. The prescriptive model is translated into a technical prescriptive model using a set of mapping rules. Furthermore, the technical prescriptive model is translated into an optimization model. The optimization model is solved and an output defining a solution from the solved optimization model is presented.

Abstract: Embodiments of the invention build models to predict the likelihood of entities that operate in a given identifier space also operating in a disjoined identifier space based on a source panel of entities that operate in one or both of the identifier spaces. In operation, a model building engine builds a model based on features associated with the source panel and features associated with standard populations in the given identifier space. The model is used to determine whether the target entity is more similar to those entities in the source panel that operate only in the given identifier space or those entities in the source panel that operate in both identifier spaces.

Abstract: Systems and methods analyze and correct the vulnerability of individual nodes in a neural network to changes in the input data. The analysis comprises first changing the activation function of one or more nodes to make them more vulnerable. The vulnerability is then measured based on a norm on the vector of partial derivatives of the network objective evaluated on each training data item. The system is made less vulnerable by splitting the data based on the sign of the partial derivative of the network objective with respect to a vulnerable and training new ensemble members on selected subsets from the data split.

Abstract: At a network-accessible artificial intelligence service for generating content-based recommendations based on multi-perspective learned descriptors, text sections associated with a plurality of description perspectives, including a single-character perspective and a multi-character perspective, are extracted from various text sources. Using the text sections as input, a machine learning model which includes respective portions corresponding to the different perspectives is trained to reconstruct the input using intermediary descriptors learned from the input. An indication that a second text source is recommended with respect to a first text source is generated using a set of the learned descriptors and transmitted.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for performing a machine learning task on a network input to generate a network output. In one aspect, one of the systems includes an attention neural network configured to perform the machine learning task, the attention neural network including one or more LSH attention layers, each LSH attention layer comprising one or more LSH attention sub-layers, each LSH sub-layer configured to: receive a sequence of queries derived from an input sequence to the LSH attention layer, the sequence of queries having a respective query at each of a plurality of input positions; determine one or more respective hash values for each of the respective queries at each of the plurality of input positions; generate a plurality of LSH groupings; and generate an attended input sequence.

Abstract: Methods and systems for generating a multi-model ensemble of global climate simulation data from a plurality of global climate simulation models (GCMs), to be used in training a neural network (NN)-based climate forecasting model, are disclosed. The methods and systems perform steps of computing a GCM validation measure for each GCM; selecting a validated subset of the GCMs, by comparing each computed GCM validation measure to a validation threshold determined based on observational historical climate data; computing a forecast skill score for each validated GCM, based on a first forecast function; selecting a validated and skillful subset of GCMs; generating one or more candidate ensembles by combining simulation data from at least two validated and skillful GCMs; computing an ensemble forecast skill score for each candidate ensemble, based on a second forecast function; and selecting a best-scored candidate ensemble.