Abstract: A computer-implemented method is provided for model training performed by a processing system. The method comprises determining a set of first weights based on a first matrix associated with a source model, determining a set of second weights based on the set of first weights, forming a second matrix associated with a target model based on the set of first weights and the set of second weights, initializing the target model based on the second matrix, and training the target model.

Abstract: This disclosure relates to the field of artificial intelligence, and discloses a data processing method. The method includes: obtaining a transformer model including a target network layer and a target module; and processing to-be-processed data by using the transformer model, to obtain a data processing result. The target module is configured to: perform a target operation on a feature map output at the target network layer, to obtain an operation result, and fuse the operation result and the feature map output, to obtain an updated feature map output. In this disclosure, the target module is inserted into the transformer model, and the operation result generated by the target module and an input are fused, so that information carried in a feature map output by the target network layer of the transformer model is increased.

Abstract: A method is provided for quantizing a neural network model performed by a processing system. The method comprises determining a scaling factor based on a distribution of weights associated with the neural network model, determining quantized weights based on the scaling factor and the weights associated with the distribution, determining a training loss of the neural network model based on the quantized weights during training of the neural network model, and determining an updated scaling factor for the neural network model based on a gradient of the training loss.

Abstract: A model compression method is provided, which can be applied to the field of artificial intelligence. The method includes: obtaining a first neural network model, a second neural network model, and a third neural network model; processing first to-be-processed data using the first neural network model, to obtain a first output; processing the first to-be-processed data using the third neural network model, to obtain a second output; determining a first target loss based on the first output and the second output, and updating the second neural network model based on the first target loss, to obtain an updated second neural network model; and compressing the updated second neural network model to obtain a target neural network model. The model generated based on the method has higher processing precision.

Abstract: A paraphrase sentence generation method and apparatus relating to the research field of natural language processing include generating m second sentences based on a first sentence and a paraphrase generation model, determining a matching degree between each of the m second sentences and the first sentence based on a paraphrase matching model, and determining n second sentences from the m second sentences based on matching degrees among the m second sentences and the first sentence, where the paraphrase generation model is obtained through reinforcement learning-based training based on a reward of the paraphrase matching model.

Abstract: In a data processing method, a processing device obtains a first neural network model and an available resource state of a terminal device, and determines a second neural network model based on the first neural network model and the available resource state. An appropriate model size is determined based on the available resource state, and a part of the first neural network model is selected, based on the determined model size, as the second neural network model on which data processing is to be performed.

Abstract: Embodiments of the present disclosure disclose a command execution method and apparatus, a terminal, and a server related speech recognition and natural language processing. In the command execution method, during an interaction between a terminal and a user, a server configured to execute a user command or the terminal may store slots and GUI information corresponding to the slots. When the filling information of the slots configured for the user command is missing, the server configured to execute the user command may obtain the missing filling information of the slots from the stored GUI information, to avoid multiple interactions between the user and the terminal. The interaction between the user and the terminal is made more intelligent, thus improving command execution efficiency.

Abstract: A text processing model training method, and a text processing method and apparatus in the natural language processing field in the artificial intelligence field are disclosed. The training method includes: obtaining training text; separately inputting the training text into a teacher model and a student model to obtain sample data output by the teacher model and prediction data output by the student model; the sample data includes a sample semantic feature and a sample label; the prediction data includes a prediction semantic feature and a prediction label; and the teacher model is a pre-trained language model used for text classification; and training a model parameter of the student model based on the sample data and the prediction data, to obtain a target student model. The method enables the student model to effectively perform knowledge transfer, thereby improving accuracy of a text processing result of the student model.

Abstract: A method includes: obtaining a text entered by a user; determining at least one topic related to the text; determining a target dialogue robot from the plurality of dialogue robots based on the at least one topic related to the text and a predefined mapping relationship between a dialogue robot and a topic, where a target topic corresponding to the target dialogue robot is some or all of the at least one topic related to the text; allocating the text to the target dialogue robot; and obtaining a reply for the text from the target dialogue robot, where the reply is generated by the target dialogue robot based on at least one semantic understanding of the text.

Abstract: An apparatus is pre-equipped with a plurality of dialogue robots, and each dialogue robot is configured to conduct a human-computer dialogue based on at least one topic. The method includes: obtaining a text entered by a user; determining at least one topic related to the text, and determining a target dialogue robot from the plurality of dialogue robots based on the at least one topic related to the text and a predefined mapping relationship between a dialogue robot and a topic, where a target topic corresponding to the target dialogue robot is some or all of the at least one topic related to the text; and allocating the text to the target dialogue robot and obtaining a reply for the text from the target dialogue robot, where the reply is generated by the target dialogue robot based on at least one semantic understanding of the text.

Abstract: A sequence conversion method includes receiving a source sequence, converting the source sequence into a source vector representation sequence, obtaining at least two candidate target sequences and a translation probability value of each of the at least two candidate target sequences according to the source vector representation sequence, adjusting the translation probability value of each candidate target sequence, selecting an output target sequence from the at least two candidate target sequences according to an adjusted translation probability value of each candidate target sequence, and outputting the output target sequence. Hence, loyalty of a target sequence to a source sequence can be improved during sequence conversion.

Abstract: An information processing method includes receiving first request information entered by a user, determining a first task engine for the first request information, where a first slot is set in the first task engine, extracting key information from the first request information based on the first slot, and if the key information fails to be extracted from the first request information based on the first slot, or if the key information is extracted from the first request information based on the first slot, but the extracted key information does not meet a condition, obtaining target key information from a shared parameter list of the user.

Abstract: A paraphrase sentence generation method and apparatus relating to the research field of natural language processing include generating m second sentences based on a first sentence and a paraphrase generation model, determining a matching degree between each of the m second sentences and the first sentence based on a paraphrase matching model, and determining n second sentences from the m second sentences based on matching degrees among the m second sentences and the first sentence, where the paraphrase generation model is obtained through reinforcement learning-based training based on a reward of the paraphrase matching model.

Abstract: An apparatus is pre-equipped with a plurality of dialogue robots, and each dialogue robot is configured to conduct a human-computer dialogue based on at least one topic. The method includes: obtaining a text entered by a user; determining at least one topic related to the text, and determining a target dialogue robot from the plurality of dialogue robots based on the at least one topic related to the text and a predefined mapping relationship between a dialogue robot and a topic, where a target topic corresponding to the target dialogue robot is some or all of the at least one topic related to the text; and allocating the text to the target dialogue robot and obtaining a reply for the text from the target dialogue robot, where the reply is generated by the target dialogue robot based on at least one semantic understanding of the text.

Abstract: A sequence conversion method includes receiving a source sequence, converting the source sequence into a source vector representation sequence, obtaining at least two candidate target sequences and a translation probability value of each of the at least two candidate target sequences according to the source vector representation sequence, adjusting the translation probability value of each candidate target sequence, selecting an output target sequence from the at least two candidate target sequences according to an adjusted translation probability value of each candidate target sequence, and outputting the output target sequence. Hence, loyalty of a target sequence to a source sequence can be improved during sequence conversion.

Abstract: A method and apparatus for determining key social information, comprises acquiring directly-retransmitted social information and indirectly-retransmitted social information of original social information, and establishing a social information retransmitting tree; acquiring an information characteristic of each piece of retransmitted social information in the social information retransmitting tree; determining a characteristic vector of each piece of retransmitted social information according to the information characteristic of the retransmitted social information; inputting the obtained characteristic vector into a preset filtering model, and acquiring candidate key social information; and selecting final key social information from all candidate key social information according to a criticality evaluation value of each piece of candidate key social information.

Abstract: A similarity of a first video to a second video may be identified automatically. Images are received from the videos, and divided into sub-images. The sub-images are evaluated based on a feature common to each of the sub-images. Binary representations of the images may be created based on the evaluation of the sub-images. A similarity of the first video to the second video may be determined based on a number of occurrences of a binary representation in the first video and the second video.

Abstract: A similarity of a first video to a second video may be identified automatically. Images are received from the videos, and divided into sub-images. The sub-images are evaluated based on a feature common to each of the sub-images. Binary representations of the images may be created based on the evaluation of the sub-images. A similarity of the first video to the second video may be determined based on a number of occurrences of a binary representation in the first video and the second video.

Abstract: A similarity of a first video to a second video may be identified automatically. Images are received from the videos, and divided into sub-images. The sub-images are evaluated based on a feature common to each of the sub-images. Binary representations of the images may be created based on the evaluation of the sub-images. A similarity of the first video to the second video may be determined based on a number of occurrences of a binary representation in the first video and the second video.