33results about How to "Reduce training requirements" patented technology

The present invention is directed to techniques for providing retail customers a seamless, individualized discovery and shopping experience. With an improved shopper / customer accounting system, along with other technologies such as intelligent fixtures, mobile devices, and product simulations, a shopper's product discovery, purchase, and post-sale experience can be highly integrated and personalized. For example, the shopper's out-of-store (online) and in-store product discovery sessions may be seamlessly linked and analyzed to provide the shopper with customized recommendations and assistance. Other variations and enhancements are disclosed.

Control systems and methods of use provide for fully automated phases of flight of an aircraft. Such fully automated phases include takeoff, cruising flight, and landing without the need for operator input or other operator intervention. Control systems and methods also provide for self-limited compliance with an operators desired deviation from a predetermined flight path, as well as automatic contingency response to non-normal conditions. Onboard and / or ground-based operators may cooperate with the control system in order to control the associated aircraft. Furthermore, an operator need not have any flight skill in order to affect changes in the flight path or other aspects of flight control.

The invention belongs to the field of hyper-spectral intelligent perception, and discloses a hyper-spectral unknown category target detection method based on a probability model and deep learning, which comprises the following steps: S1, inputting hyper-spectral training data into a trained CNN classification model, and outputting activation vectors of all samples of each category; S2, accumulating and averaging the activation vectors of all the samples which belong to the same category and are classified correctly to obtain an average activation vector, and representing the center of the category by using the average activation vector; S3, fitting a Weibull model belonging to each category based on the activation vectors of all the samples in each category and the mean activation vector of the category; S4, inputting hyperspectral test data into the CNN model and the probability model to form a network based on the Weibull fitting result of each category, and calculating the probability belonging to an unknown category; The method is clear in structure and easy to implement, the training requirement of a neural network learning model is lowered, and the effect of unknown categorytarget detection can be obviously improved.

A method of acquiring at least one clinical MRI image of a subject comprising the following steps: acquiring a first survey image with a first field of view, the first survey image having a first spatial resolution,—locating a first region of interest and at least one anatomical landmarks in the first survey image, determining the position and the orientation of the first region of interest using the anatomical landmarks, the position and the orientation of the first region being used for—planning a second survey image,—acquiring the second survey image with a second field of view, the second survey image having a second spatial resolution, the second spatial resolution being higher than the first spatial resolution, generating a geometry planning for the anatomical region of interest using the second survey image,—acquiring a diagnostic image of the anatomical region of interest using the geometry planning.

The present invention relates to a system and methodology to facilitate machine learning and predictive capabilities in a processing environment. In one aspect of the present invention, a Mutual Information Model is provided to facilitate predictive state determinations in accordance with signal or data analysis, and to mitigate classification error. The model parameters are computed by maximizing a convex combination of the mutual information between hidden states and the observations and the joint likelihood of states and observations in training data. Once the model parameters have been learned, new data can be accurately classified.

A method and system for traffic control rule generation are provided. Configuration information for the communication device, default information stored for the communication device, or both are accessed. One or more parameters affecting the processing of communication traffic by the communication device are determined from the accessed information, and the one or more parameters are used to generate communication traffic control rules to be applied to communication traffic at the communication device. The generated traffic control rules are applied at the interface of the communication device to traffic terminated by the communication device.

An assistive device for subcutaneous injections or implants. The assistive device is placed under a blood pressure cuff that displaces the skin and subcutaneous tissues of a patient at an insertion site in order to allow guided insertion of a needle or implant applicator into a subcutaneous fat layer. The skin and subcutaneous tissue is displaced or drawn, via pressure applied by the blood pressure cuff, through an injection window of a device body into a hollow cavity of the device body. A guide channel leads from a front side of the device body into the hollow cavity, and an applicator of an adapter is inserted through the guide channel into the subcutaneous fat layer.

The present invention is directed to techniques for providing retail customers a seamless, individualized discovery and shopping experience. With an improved shopper / customer accounting system, along with other technologies such as intelligent fixtures, mobile devices, and product simulations, a shopper's product discovery, purchase, and post-sale experience can be highly integrated and personalized. For example, the shopper's out-of-store (online) and in-store product discovery sessions may be seamlessly linked and analyzed to provide the shopper with customized recommendations and assistance. Other variations and enhancements are disclosed.

The invention discloses an intelligent transformer station mixing simulation training system, and belongs to a training reaching system. The system provided by the invention comprises an electrical network digital simulation system, a primary equipment simulation device, and secondary equipment of an intelligent transformer station. The electrical network digital simulation system comprises an electrical network real-time simulation server, a faculty machine and an I / O interface; the primary equipment simulation device comprises an electronic mutual inductor simulation device, a low-voltage switch simulation device, a breaker simulation device and a knife simulation device; the electrical network digital simulation system is connected with the primary equipment simulation device; and the primary equipment simulation device is connected with the secondary equipment. According to the invention, an integrated training platform is established for professional training of an intelligent electrical network, so that full-scope full-process and full-scene simulation training for transformer station operation personnel can be realized; an open principle is observed in terms of a network structure and hardware configuration, so that expansion capability and maintainability of the system are higher; and the network configuration and functions exhibit certain foresight, so that the needs for further development and training are satisfied, and a good base is laid for later function upgrading.

The invention provides a penalty dual decomposition channel decoding method assisted by a multi-layer neural network, which is mainly oriented to binary linear codes. The method includes the following steps: (1) Constructing the maximum likelihood optimization problem based on channel decoding; (2) Transforming the maximum likelihood optimization problem of channel decoding into parity-check polyhedron-based decoding by introducing the concept of basic polyhedron into constraints code optimization problem; (3) introduce the solution penalty dual decomposition method to solve the decoding optimization problem, and obtain the penalty dual decomposition channel decoder; (4) design a check polyhedron map based on a multi-layer neural network, and obtain learning parameters through training, The multi-layer neural network-based check polyhedron mapping is introduced into the penalty dual decomposition channel decoder, and the multi-layer neural network-assisted penalty dual decomposition channel decoder is obtained. The present invention further improves the decoding performance and reduces the decoding delay by means of the power of machine learning.

A train control system, especially a train control system for train formation using Distributed Power Supply (DP) technology. The technique involves the positioning and operation of one or more sets of locomotives distributed throughout a train consist comprising multiple railcars and multiple locomotives. These locomotives are controlled remotely from the cab in the lead locomotive, ie the lead locomotive (LL).

An assistive device for subcutaneous injections or implants. The assistive device is placed under a blood pressure cuff that displaces the skin and subcutaneous tissues of a patient at an insertion site in order to allow guided insertion of a needle or implant applicator into a subcutaneous fat layer. The skin and subcutaneous tissue is displaced or drawn, via pressure applied by the blood pressure cuff, through an injection window of a device body into a hollow cavity of the device body. A guide channel leads from a front side of the device body into the hollow cavity, and an applicator of an adapter is inserted through the guide channel into the subcutaneous fat layer.

The invention discloses a content-based matching recommendation method for user personalized products, which overcomes the problem that the recommendation algorithm in the prior art still needs to be improved. The invention includes step 1, user-based random batch sampling method; step 2, content-based user-product matching method: establish a network; according to batch input based on user-based random batch sampling method, orderly user ids are obtained, and a list of product ids is recorded in history , a set of target product id and label, train, adjust, and evaluate the network model on the training set, verification set, and test set respectively; input a specific user and its history, and use the content-based user-product matching network to predict its response to all unwatched Score and sort the movies, and finally output top-N recommendation results. The invention adopts a lightweight neural network, which greatly reduces the training time and training equipment requirements, the sampling process is easy, the model input is more random, and the generalization ability of the prediction result is stronger.

The invention discloses an ice hockey training device, which comprises a goal board, at least one goal return pipe is connected to the goal board, the ball enters from one end of the goal return pipe, and the ball enters from the other end of the goal return pipe. One end discharges, this device can realize the automatic return of the ball without power supply, this device can be arranged at any position on the ice surface and any position on the land for training, which greatly reduces the training cost and the requirements of the training site.

The invention belongs to the field of hyper-spectral intelligent perception, and discloses a hyper-spectral unknown category target detection method based on a probability model and deep learning, which comprises the following steps: S1, inputting hyper-spectral training data into a trained CNN classification model, and outputting activation vectors of all samples of each category; S2, accumulating and averaging the activation vectors of all the samples which belong to the same category and are classified correctly to obtain an average activation vector, and representing the center of the category by using the average activation vector; S3, fitting a Weibull model belonging to each category based on the activation vectors of all the samples in each category and the mean activation vector of the category; S4, inputting hyperspectral test data into the CNN model and the probability model to form a network based on the Weibull fitting result of each category, and calculating the probability belonging to an unknown category; The method is clear in structure and easy to implement, the training requirement of a neural network learning model is lowered, and the effect of unknown categorytarget detection can be obviously improved.

A system and method of tracking location and orientation of power tools utilized in the assembly and maintenance of complex systems is disclosed. The system can facilitate collaboration between maintenance and alert maintenance personnel to complete complex system activities.