Abstract: Described herein are embodiments of a reinforcement learning based large-scale multi-objective ranking system. Embodiments of the system may be used for optimizing short-video recommendation on a video sharing platform. Multiple competing ranking objective and implicit selection bias in user feedback are the main challenges in real-world platform. In order to address those challenges, multi-gate mixture of experts (MMoE) and soft actor critic (SAC) are integrated together into a MMoE_SAC system. Experiment results demonstrate that embodiments of the MMoE_SAC system may greatly reduce a loss function compared to systems only based on single strategies.

Abstract: The present disclosure provides a method and an apparatus for analyzing an intention based on artificial intelligence. The method includes: receiving a query; acquiring a keyword of an intention of the query according to a preset strategy; acquiring a qualifier of the intention of the query according to the keyword based on a syntax tree; and determining the intention of the query according to the keyword and the qualifier.

Abstract: Embodiments of the present disclosure provide a method and apparatus for recommending a test question, and an intelligent device. The method includes: acquiring a plurality of skill entities of a post; calculating, according to the data of the post, a weight value of each of the plurality of skill entities; and acquiring, according to the weight value of each skill entity, a recommended test question of the post from a question bank.

Abstract: A method and an apparatus for generating an image are provided. A specific embodiment of the method comprises: acquiring a to-be-processed facial image, the image resolution of the to-be-processed facial image being lower than a preset first resolution threshold; and inputting the to-be-processed facial image into a pre-trained generativemodel to generate a processed facial image. The generative model updates a model parameter using a loss function in a training process, and the loss function is determined based on a probability of an image group being positive sample data, the image group composed of a facial sample image and a facial generative image. According to this embodiment, authenticity of the generated facial image is enhanced.

Abstract: The present disclosure provides a battery detection method and a battery detection device. The method includes: obtaining a picture of each battery on a battery production line, and obtaining a corresponding production node; inputting the picture into a preset defect detection model, and obtaining a detection result output by the defect detection model, and when the detection result denotes that there is the defect on the picture, sending a control instruction to a control device of the production node corresponding to the picture, to cause the control device to shunt the battery corresponding to the picture having the defect based on the control instruction. The detection result includes whether there is a defect, a defect type, and a defect position.

Abstract: A method and device for correcting an error in a text are provided. The method includes: preprocessing the text to obtain at least one segment of the text; generating a plurality of candidate segments for the segment; scoring the plurality of candidate segments with a tree model, to obtain respective first scoring results of the plurality of candidate segments; scoring the plurality of candidate segments with a deep neural network model, to obtain respective second scoring results of the plurality of candidate segments; for each candidate segment, calculating a scoring of the candidate segment based on the first scoring result and the second scoring result of the candidate segment; ranking the plurality of candidate segments according to the scorings of the candidate segments, to obtain a ranking result; and correcting the error in the text according to the ranking result.

Abstract: Provided are a video pushing method, an apparatus, a device, and a storage medium. In an embodiment of the present disclosure, the number of times a user page is displayed on a user terminal is obtained; if the number of times is less than a preset value, a target video to be pushed to the user terminal is determined according to a popularity of a video in a storage system corresponding to a server; if the number of times is greater than a preset value, a target video to be pushed to the user terminal is determined according to feature information of a user and feature information of a video in the storage system corresponding to the server. Accordingly, the server can continuously push videos that meet the user's needs to the user terminal while the user is using the user page, thereby improving the user experience.

Abstract: The present application provides a model training, image processing method, device, storage medium, and program product relating to deep learning technology, which are able to screen auxiliary image data with image data for learning a target task, and further fuse the target image data and the auxiliary image data, so as to train a built and to-be-trained model with the fusion-processed fused image data. This implementation can increase the amount of data for training the model, and the data for training the model is determined is based on the target image data, which is suitable for learning the target task. Therefore, the solution provided by the present application can train an accurate target model even if the amount of target image data is not sufficient.

Abstract: The present disclosure provides a deep learning assignment processing method and apparatus, a device and a storage medium. It is feasible to obtain the deep learning assignment submitted by the user in a predetermined manner, the predetermined manner comprising the web UI manner, then submit the deep learning assignment to the deep learning system so that the deep learning system runs the submitted deep learning assignment. As compared with the prior art, processing such as programming is not needed upon submitting the deep learning assignment in the solutions of the present disclosure, thereby simplifying the user's operations, improving the processing efficiency of the deep learning assignment, and accelerating the user's speed of developing deep learning.

Abstract: Described herein are systems and methods to prune deep neural network models in reducing the overall memory and compute requirements of these models. It is demonstrated that using block pruning and group lasso combined with pruning during training, block-sparse recurrent neural networks (RNNs) may be built as accurate as dense baseline models. Two different approaches are disclosed to induce block sparsity in neural network models: pruning blocks of weights in a layer and using group lasso regularization to create blocks of weights with zeros. Using these techniques, it is demonstrated that block-sparse RNNs with high sparsity can be created with small loss in accuracy. Block-sparse RNNs eliminate overheads related to data storage and irregular memory accesses while increasing hardware efficiency compared to unstructured sparsity.

Abstract: Provided are a method for processing automobile image data, an apparatus, and a readable storage medium. The method includes training a preset adversarial network according to automobile image sample data obtained by collecting and preset random noise information, to obtain a trained adversarial network for generating automobile image data; and inputting the automobile image sample data into the trained adversarial network, and generating the automobile image data including a variety of the random noise information, according to a preset generation target, wherein the automobile image data are configured to train an automobile image recognition neural network to perform automobile image recognition. Therefore, the automobile image data that may be used to train an automobile image recognition neural network are generated by using the adversarial network, which amplifies the amount of data, and reduces the cost of obtaining data.

Abstract: Embodiments of the present disclosure provides a method and apparatus for inspecting burrs of an electrode slice. The method may include: acquiring a to-be-inspected electrode slice image; and inputting the to-be-inspected electrode slice image into a pre-trained burr instance segmentation model to obtain inspection result for characterizing whether the electrode slice displayed in the to-be-inspected electrode slice image has burrs and contour of the burrs, where the burr instance segmentation model is used to characterize the corresponding relationship between the electrode slice image and the inspection result and contour information. The method may further include: and outputting, in response to the inspection result for characterizing that the electrode slice displayed in the to-be-inspected electrode slice image has burrs, prompt information for characterizing that the electrode slice displayed in the to-be-inspected electrode slice image has burrs.

Abstract: The present disclosure proposes a language generation method and apparatus. The method includes: performing encoding processing on an input sequence by using a preset encoder to generate a hidden state vector corresponding to the input sequence; in response to a granularity category of a second target segment being a phrase, decoding a first target segment vector, the hidden state vector, and a position vector corresponding to the second target segment by using N decoders to generate N second target segments; determining a loss value based on differences between respective N second target segments and a second target annotated segment; and performing parameter updating on the preset encoder, a preset classifier, and the N decoders based on the loss value to generate an updated language generation model for performing language generation.

Abstract: A resource configuration method and apparatus for heterogeneous cloud services are provided. The method may include: establishing a basic model with a general structure for at least two heterogeneous cloud services, where the basic model comprises a trend model and a periodic model; determining a cloud service in the at least two cloud services as a target cloud service, and acquiring a target historical data set of the target cloud service; training the trend model and the periodic model using the target historical data set; generating a target prediction model corresponding to the target cloud service based on the trained trend model and the trained periodic model; and generating, based on the target prediction model, a resource amount demanded by the target cloud service in a future time period, and configuring resources for the target cloud service according to the demanded resource amount.

Abstract: A road condition predicting method and apparatus, a computer device and a readable medium. The method comprises: obtaining N-order neighboring roads of a target road in a road network and obtaining feature information of the target road at a current time instant and feature information of N-order neighboring roads of the target road. The feature information including road conditions at the current and at at least one historical time; predicting road conditions of the target road at a preset future time according to the feature information of the target road at the current time, the feature information of the N-order neighboring roads and a pre-trained road condition predicting model. Reference is made to spatial information of the target road, namely, and to the road condition of the N-order neighboring roads of the target road at the current time instant and the road conditions at at least one historical time instant.

Abstract: A system comprises: a data warehouse, a storage device and a cluster including a plurality of computing nodes; the data warehouse is configured to store task data obtained from the user; at least one computing node in the cluster includes a resource scheduling component, and is configured to perform resource scheduling for the task and determine a computing node executing the task; the computing node executing the task comprises a model training component and/or a prediction component; the model training component is configured to, according to task data, invoke a corresponding type of learning model from the storage device; use sample data and training target included in the task data to train the learning model, to obtain the prediction model corresponding to the task and store the prediction model in the storage device; the prediction component is configured to obtain a prediction result output by the prediction model.