Abstract: Techniques for identifying fraudulent transactions are described. In one example method, an operation sequence and time difference information associated with a transaction are identified by a server. A probability that the transaction is a fraudulent transaction is predicted based on a result provided by a deep learning network, where the deep learning network is trained to predict fraudulent transactions based on operation sequences and time differences associated with a plurality of transaction samples, and where the deep learning network provides the result in response to input including the operation sequence and the time difference information associated with the transaction.

Abstract: Systems, apparatus and methods are described including operations for a flexible neural network accelerator.

Abstract: The present disclosure provides a method and a system for optimizing Hidden Markov Model based land change prediction. Firstly, remotely sensed data is pre-processed and classified into a plurality of land use land cover classes (LULC). Then socio-economic driver variables data for a pre-defined interval of time are provided from a database. A Hidden Markov Model (HMM) is defined with LULC as hidden states and socio-economic driver variables data as observations and trained for generating a MINI state transition probability matrix. Again the defined MINI is trained by taking input data from scenario based temporal variables to generate another set of HMM state transition probability matrix. The generated MINI state transition probability matrix is then integrated with a spatio-temporal model to obtain an integrated model for predicting LULC changes to generate at least one prediction image.

Abstract: An approach is provided for an information handling system that includes a processor and a memory to improve the quality of question-answer sets used as inputs to a question-answering (QA) system. In the approach, a question-answer pair is analyzed using natural language processing (NLP) components. Some of the NLP components may be taken from the QA system whose input is being analyzed The question-answer pair includes a question and an answer to the question. Based on the analysis, one or more shortcomings of the question-answer pair are identified. The shortcomings relate to an ability of the target QA system to analyze the question. A human-readable feedback is provided to a user. The feedback recommends one or more possible actions to address the identified shortcomings.

Abstract: A computing device receives training data representing different observations where each observation is categorized into one of options for a target variable. The device obtains computer command(s) for categorizing into one of the options for the target variable. The device generates a sampling scheme for sampling terms of the training data. The device generates sampling models by, for N iterations of the sampling scheme: determining a subset of the training data based on a training data index; sampling, based on a term index, the subset of the training data for a subset of terms; and generating, based on the subset of terms, a sampling model for categorizing, according to the computer command(s). Each sampling model is generated from a different subset of terms such that the sampling models are randomized. The device computes an aggregated model for categorizing test data into one of the options for the target variable.

Abstract: A method, system, and computer program product, include identifying a plurality of pollution process sets and determining pollution sources based on pollution start times of target pollution processes with matched features in the plurality of pollution process sets within a time window.

Abstract: An apparatus and a method. The apparatus includes a plurality of long short term memory (LSTM) networks, wherein each of the plurality of LSTM networks is at a different network layer, wherein each of the plurality of LSTM networks is configured to determine a residual function, wherein each of the plurality of LSTM networks includes an output gate to control what is provided to a subsequent LSTM network, and wherein each of the plurality of LSTM networks includes at least one highway connection to compensate for the residual function of a previous LSTM network.

Abstract: One embodiment provides a method comprising intercepting a voice communication, collecting multi-sensory inputs associated with the voice communication, and determining an overall risk assessment metric for the voice communication based on the multi-sensory inputs and learned signatures. The multi-sensory inputs are indicative of content of the voice communication and one or more contextual factors associated with a target of the voice communication. The overall risk assessment metric indicates a likelihood the voice communication is a scam.

Abstract: A system and method for training a computerized data model for the algorithmic detection of non-linearity in a data set includes providing two master data sets corresponding to two discrete time periods, respectively, and a third data set for a third discrete time period. The two master data sets are mapped to at least one code model. A stacking average model is trained with the at least two master data sets corresponding to two discrete time periods by using a stacked regression algorithm. A box-cox transformation function is applied to the models to provide a predicted value for the third data set of the third discrete time period. An ensemble is created using the predicted value for the third data set and the first, second, and third models of the trained stacking average model to identify a non-linearity in the third data set.

Abstract: A mechanism is provided in a data processing system for rating difficulty of a question. The mechanism receives an input question and generates one or more candidate answers from a corpus of knowledge using a pipeline of software engines. The pipeline of software engines generates a plurality of features extracted from the question, the one or more candidate answers, or the corpus of knowledge. The mechanism then generates a question difficulty score based on the plurality of features using a machine learning model. The machine learning model maps features to assigned weights for scaling the difficulty score.

Abstract: Technical solutions are described for visually depicting health of a cognitive system. An example computer-implemented method includes accessing a query-log of a question input to the cognitive system. The method also includes generating a query-node corresponding to the question. The method also includes configuring animation parameters of the query-node based on the query-log. The method also includes displaying the query-node according to the animation parameters.

Abstract: An apparatus and a method. The apparatus includes a first recurrent network in a first layer; a second recurrent network in a second layer connected to the first recurrent network; a distant input gate connected to the second recurrent network; a first highway gate connected to the distant input gate and the second recurrent network; a first elementwise product projection gate connected to the distant input gate, the highway gate, and the second recurrent network; a second highway gate connected to the first recurrent network and the second recurrent network; and a second elementwise product projection gate connected to the first recurrent network, the second highway gate, and the second recurrent network.