Techniques are described related to turn-based reinforcement learning for dialog management. In various implementations, dialog states and corresponding responsive actions generated during a multi-turn human-to-computer dialog session may be obtained. A plurality of turn-level training instances may be generated, each including: a given dialog state of the plurality of dialog states at an outset of a given turn of the human-to-computer dialog session; and a given responsive action that was selected based on the given dialog state. One or more of the turn-level training instances may further include a turn-level feedback value that reflects on the given responsive action selected during the given turn. A reward value may be generated based on an outcome of the human-to-computer dialog session. The dialog management policy model may be trained based on turn-level feedback values of the turn-level training instance(s) and the reward value.

Turn-based reinforcement learning for dialog management

The invention discloses a positioning method and a device and a terminal. The positioning method comprises the steps that the positioning position of the terminal is obtained through a positioning system; a view photo of the current position of the terminal is photographed through a camera; the positioning position and the view photo are sent to a positioning server, so that the positioning server calibrates the positioning position according to the view photo, and the calibrated positioning position is obtained; the calibrated positioning position sent by the positioning server is received. According to the technical scheme in the embodiment, the positioning position of the terminal is calibrated through the view photo, and more accurate positioning is achieved.

Positioning method and device and terminal

The disclosed computer-implemented method may include (i) monitoring computing activity, (ii) detecting, during a specific time period, at least one malicious network connection that involves a computing device within a network, (iii) determining that no malicious network connections involving the computing device were detected during another time period, (iv) identifying a feature of the computing activity that (a) occurred during the specific time period and (b) did not occur during the other time period, (v) determining that the feature is likely indicative of malicious network activity due at least in part to the feature having occurred during the specific time period and not having occurred during the other time period, and in response to detecting the feature at a subsequent point in time, (vi) performing a security action on a subsequent network connection attempted around the subsequent point in time. Various other methods, systems, and computer-readable media are also disclosed.

Systems and methods for preventing malicious network connections using correlation-based anomaly detection

The invention discloses a shopping guide map generating method and device. The method comprises performing dynamic modelling for multiple pictures of space to be disposed to obtain the geographic coordinates of the shelf in the space object to be disposed; determining the geographic coordinates of the wireless equipment installed on the shelf in accordance with the geographic coordinates of the shelf and the correspondence between the shelf and the wireless equipment; locating geographic coordinates of object on the shelf in accordance with the geographic coordinates of the wireless equipment; generating shopping guide map of the space object to be disposed in accordance with the geographic coordinates of the shelf, the physical object and the wireless equipment. By the method, efficiency in generating the shopping guide map as well as accuracy of the shopping guide map can be improved.

Shopping guide map generating method and device

A system includes a memory and a processor configured to analyze a data set to determine a first number of incidents during a first period of time and a second number of incidents during a second period of time, train a plurality of models predict a number of incidents during the second period of time, wherein the plurality of respective models comprise a random forest model, a drift model, and a naive seasonal drift model, identifying the model that best predicted the number of incidents during the second period of time, and utilizing the identified model to predict a third number of incidents within a set allowable range of values during a third period of time, and upper and lower limits of the third number of incidents during the third period of time based on a set confidence level and displaying the third number of incidents during the third period of time.

Systems and method for incident forecasting

A method for predicting anomalies in a computer application includes during runtime of the computer application, detecting traffic metrics and incident tickets associated with the computer application, the incident ticket indicating an incident might occur in the computer application; calculating a threshold based on absolute values of second order differences associated with the traffic metrics, wherein the threshold is such that when the absolute value of the second order difference associated with the traffic metrics exceeds the threshold, a recall rate R recall that the computer application is recalled is maximized; obtaining predicted metrics of the computer application in a next time period based on the traffic metrics; and in response to an absolute value of a second order difference associated with the predicted metrics exceeding the threshold, predicting potential anomalies of the computer application in the next time period.

Predicting anomalies and incidents in a computer application

A method photo-based positioning includes obtaining a positioning photo taken by a mobile device within a predetermined space; comparing multiple feature elements exacted from sampling photos taken within the predetermined space in advance with the positioning photo to determine each feature element existing in the positioning photo; obtaining a first position coordinate of each feature element which is determined to exist in the positioning photo in a sampling photo corresponding to the positioning photo, and a second position coordinate of each feature element which is determined to exist in the positioning photo in the positioning photo; and calculating position parameters of the mobile device by using each of the obtained first position coordinate and second position coordinate, wherein the least number of feature elements existing in the positioning photo is determined according to the number of the position parameters.

Photo-based positioning

The invention discloses a positioning method and apparatus based on photographs. The method comprises the steps of: obtaining a positioning photograph shot by a mobile device in a reserved space; contrasting a plurality of characteristic elements extracted in advance from a sampling photograph shot in the reserved space with the positioning photograph to determine each characteristic element existing in the positioning photograph; obtaining a first position coordinate of each characteristic element determined to exist in the positioning photograph in a sampling photograph corresponding with the positioning photograph, and a second position coordinate in the positioning photograph; and utilizing each obtained first position coordinate and second first position coordinate to calculate position parameters of the mobile device, wherein a minimum number of the characteristic elements in the positioning photograph is determined according to the number of the position parameters. The method and apparatus can provide accurate positional information and directional information to be used for performing accurate navigation, shopping guide, etc.

Positioning method and apparatus based on photographs

Aspects of the present invention generally relate to a system, method, and computer program product for transforming APIs from legacy system-of-record to new system-of-engagement using complete API process logic. API candidates are generated from execution traces for transforming the API from the legacy system to a new system of record. An embodiment of the invention is first, add implementers into a legacy systems and collect the instrumenter output as execution traces. Second, rank the execution traces and find the most frequent execution traces. Third, consolidate the execution traces by merging common trace segments. The codes covered by the consolidated execution traces will be considered as candidate API components. Fourth, expand the merged execution traces and find related execution traces and check the connectivity between different nodes.

System and method for automatic api candidate generation

Procedure dialogs are improved through knowledge mining within a reinforcement learning framework. Taking an existing procedure dialog as input, a machine learning model is generated. User interactions with the machine learning model are monitored and used to update the machine learning model. The updates to the machine learning model are applied to the existing procedure dialog for review and revision by subject matter experts.

Li Jun Mei

Papers

Predicting anomalies and incidents in a computer application

Photo-based positioning

Positioning method and apparatus based on photographs

System and method for automatic api candidate generation

Procedure dialogs using reinforcement learning