Abstract: Certain aspects of the present disclosure provide techniques for performing computations on encrypted data. One example method generally includes obtaining, at a computing device, encrypted data, wherein the encrypted data is encrypted using fully homomorphic encryption and performing at least one computation on the encrypted data while the encrypted data remains encrypted. The method further includes identifying a clear data operation to perform on the encrypted data and transmitting, from the computing device to a server, a request to perform the clear data operation on the encrypted data, wherein the request includes the encrypted data. The method further includes receiving, at the computing device in response to the request, encrypted output from the server, wherein the encrypted output is of the same size and the same format for all encrypted data transmitted to the server.

Abstract: Certain aspects of the present disclosure provide techniques for performing finite rank deep kernel learning. In one example, a method for performing finite rank deep kernel learning includes receiving a training dataset; forming a set of embeddings by subjecting the training data set to a deep neural network; forming, from the set of embeddings, a plurality of dot kernels; combining the plurality of dot kernels to form a composite kernel for a Gaussian process; receiving live data from an application; and predicting a plurality of values and a plurality of uncertainties associated with the plurality of values simultaneously using the composite kernel.

Abstract: A method involves receiving a first command. The first command includes a data extraction expression applied to fields of a dataset of a data source. The first command also includes a first report configuration expression applied to first dimensions of a first report. The method also involves generating, by executing the data extraction expression on the dataset, records of the dataset. The method also involves generating, by executing the first report configuration expression on the records, a first tree of subsets of the records. The method also involves populating, using the first report configuration expression and the first tree of subsets, cells of the first dimensions to obtain first populated dimensions. The method also involves generating, in response to receiving the first command and by traversing the first tree of subsets, the first report including the first populated dimensions.

Abstract: This disclosure relates to identifying anomalies in, predicting data points for, and determining a feature's importance to input time series data and outputs from the data. An example system is configured to perform operations including obtaining, by an autoencoder, time series data including multiple sequences of data points, encoding, by an encoder of the autoencoder, the obtained time series data into encoded data, decoding, by a decoder of the autoencoder, the encoded data into decoded data, reconstructing time series data from the decoded data, determining a reconstruction error based on the reconstructed time series data and the obtained time series data, identifying an anomaly based on the reconstruction error. The system is also configured to predict one or more data points from the encoded data and determine a contribution (SHAP value) of a feature to the obtained time series data that is associated with a plurality of features.

Abstract: A method and system facilitates importation and categorization of financial documents in a financial system. The method and system include receiving source data and externally generated metadata associated with a financial document, such as a receipt of purchase. The method and system further include analyzing the metadata and categorizing the financial document based on the metadata.

Abstract: Embodiments disclosed herein select a content message to present to a user on a page of an application based on paralinguistic features of audio input received from the user for the application. The audio input is received via a microphone associated with a computing device. A feature extractor extracts paralinguistic features from the audio input. A predictive model determines a label indicating a measure of receptiveness to product placement (e.g., a predicted marketing outcome) based on the paralinguistic features. A content-selection component selects a content message to present to the user based on the label and based on a profile of the user.

Abstract: Systems and methods for identifying businesses having complementary business cycles are disclosed. An example method may include receiving financial information for each business of a plurality of businesses, the financial information including a plurality of values of a financial indicator with respect to time, training a machine learning model to determine a preference metric for a pair of businesses based at least in part on respective values of the businesses' financial indicators, for each given business in the plurality of businesses, determining corresponding values of the preference metric, using the training machine learning model, for at least a respective subset of the plurality of businesses, and determining an optimal pairing for each business in the plurality of businesses based at least in part on the determined values of the preference metric.

Abstract: A method and system assist users of an electronic bookkeeping system to sort financial transactions into accounts representing bookkeeping categories. The method and system generate vectors representing the accounts and the merchants known to the bookkeeping system. The method and system generate compressed versions of the vectors by compressing the vectors. The method and system assist users to sort the financial transactions by analyzing the compressed vectors.

Abstract: A processor may generate a plurality of intermediate feature layers of an image using convolutional neural network (CNN) processing. For each intermediate feature layer, the processor may generate a plurality of text proposals using a region proposal network (RPN). Each text proposal may comprise a portion of the intermediate feature layer that is predicted to contain text. The processor may perform OCR processing on image data within a plurality of regions of the image to generate a text result for each region. Each region may comprise at least one of the text proposals. The processor may assemble the text results into a text string comprising the text results ordered according to a spatial order in which the plurality of regions appear within the image.

Abstract: This disclosure relates to predictions based on a Bernoulli uncertainty characterization used in selecting between different prediction models. An example system is configured to perform operations including determining a prediction by a first prediction model. The first prediction model is associated with a loss function. The system is also configured to determine whether the prediction is associated with the first prediction model or a second prediction model based on a joint loss function. The second prediction model is associated with a likelihood function, and the joint loss function is based on the loss function and the likelihood function. The system is further configured to indicate the prediction to the user in response to determining that the prediction is associated with the first prediction model. If the prediction is associated with the second prediction model, the system may prevent indicating the prediction to the user.

Abstract: Automatic keyphrase labeling and machine learning training may include a processor extracting a plurality of keywords from at least one search query that resulted in a selection of a document appearing in a search result. For each of the plurality of keywords, the processor may determine a probability that the keyword describes the document. The processor may generate one or more keyphrases by performing processing including selecting each of the plurality of keywords having a probability greater than a predetermined threshold value for insertion into at least one of the one or more keyphrases and assembling the one or more keyphrases from the selected plurality of keywords. The processor may label the document with the keyphrase.

Abstract: This disclosure relates to predictions based on a Bernoulli uncertainty characterization used in selecting between different prediction models. An example system is configured to perform operations including determining a prediction by a first prediction model. The first prediction model is associated with a loss function. The system is also configured to determine whether the prediction is associated with the first prediction model or a second prediction model based on a joint loss function. The second prediction model is associated with a likelihood function, and the joint loss function is based on the loss function and the likelihood function. The system is further configured to indicate the prediction to the user in response to determining that the prediction is associated with the first prediction model. If the prediction is associated with the second prediction model, the system may prevent indicating the prediction to the user.

Abstract: Systems and methods for personalizing messages in a conversational chatbot are disclosed. An example method may include receiving clickstream event data corresponding to click events by users of an application, generating featurized clickstream data based at least in part on the received clickstream event data, determining one or more predicted intentions for a first user based at least in part on the featurized clickstream data, and generating one or more personalized messages for the first user based at least in part on the one or more predicted user intentions.

Abstract: This disclosure relates to forecasting when and whether an invoice is to be paid and indicating such forecasts to a user. An example system is configured to perform operations including determining, by a classification model, a first confidence as to whether an invoice is to be paid, determining, by a regression model associated with the classification model, a first time associated with a second confidence as to when the invoice is likely to be paid, and indicating, to a user, whether the invoice is to be paid based on the first confidence and the first time when the invoice is likely to be paid based on the second confidence. The regression model may include one or more gradient boosted trees to determine the first time. Different times associated with different confidences can be determined by different gradient boosted trees, with the specific tree corresponding to the associated confidence.

Abstract: Methods, systems and articles of manufacture for using one or more predictive models to predict which tax matters are relevant to a particular taxpayer during preparation of an electronic tax return. A tax return preparation system accesses taxpayer data such as personal data and/or tax data regarding the particular taxpayer. The system executes a predictive model which receives the taxpayer data as inputs to the predictive model. The predictive model generates as output(s) one or more predicted tax matters which are determined to be likely to be relevant to the taxpayer. The system may then determine tax questions to present to the user based at least in part upon the predicted tax matters determined by the predictive model.

Abstract: A method and system generate customized financial document templates in a document preparation system, according to one embodiment. The method and system receive document data from a user corresponding to a document including a plurality of custom data fields in a custom template. The method and system map the custom data fields to system data fields from a relatively small subset of data fields selected from a system data field pool based on the characteristics of the user. The method and system generate a custom form template based on the document data and the mapping of the custom data fields to system data fields.

Abstract: Certain aspects of the present disclosure provide techniques for selecting a response to a self-support query. One example method generally includes receiving an audio stream query including spoken content from a user recorded by a mobile device and determining a set of paralinguistic features from the spoken content. The method further includes estimating an emotional state of the user based on the set of paralinguistic features and identifying subject matter of the spoken content in the audio stream query. The method further includes determining two or more query responses corresponding to the subject matter to present to the user and transmitting at least one query response to the mobile device.

Abstract: Artificial intelligence, big data, and crowd sourcing techniques are utilized to efficiently and effectively determine permissions that should be granted to a party within an organization. In one example, the permissions granted to a party within an organization are determined using one or more algorithms to identify, weight, and correlate historical and current permissions to party attributes for parties within the organization and/or for similar parties in similar organizations. In one example, the activity of the party within the organization is then monitored and the permissions granted the party are automatically modified as needed to allow the party to perform their tasks in the organization as the party's responsibilities within the organization evolve.

Abstract: A method ranks image brands. An image brand model is trained to generate an image brand rank from image features. An augmented image brand model is trained to generate an augmented image brand rank from the image brand rank. Predicted financial features are generated from the augmented image brand rank using a feature generation model. A neural network model is trained to generate a predicted augmented image brand rank from the predicted financial features.

Abstract: A method including embedding, by a trained issue MLM (machine learning model), a new natural language issue statement into an issue vector. An inner product of the issue vector with an actions matrix is calculated. The actions matrix includes centroid-vectors calculated using a clustering method from a second output of a trained action MLM which embedded prior actions expressed in natural language action statements taken as a result of prior natural issue statements. Calculating the inner product results in probabilities associated with the prior actions. Each of the probabilities represents a corresponding estimate that a corresponding prior action is relevant to the issue vector. A list of proposed actions relevant to the issue vector is generated by comparing the probabilities to a threshold value and selecting a subset of the prior actions with corresponding probabilities above the threshold. The list of proposed actions is transmitted to a user device.