Abstract: A system can receive data from a device on a channel for broadband cellular communications. The system can, based on determining that a cyclic redundancy check passes for the data, decode a payload of the data to produce a transmitted signal. The system can determine a metric of signal quality for the channel based on at least a portion of the transmitted signal that is separate from a pilot resource. The system can compare the metric of signal quality to a value specified by a signal quality criterion, to produce a signal quality result, wherein, based on the signal quality result indicating that the metric of signal quality is less than the value specified by the signal quality criterion, determine that the cyclic redundancy check passing for the data comprises a false alarm, and classify the data as a discontinuous transmission; and otherwise, determine that the data is valid.

Abstract: A system can utilize a first common physical uplink control channel (PUCCH) format 1 communication format for broadband cellular communications with user equipment via a channel. The system can estimate a first channel characteristic based on performing channel estimation on the channel over uplink signals, which comprises a Doppler spread. The system can estimate a second channel characteristic based on the performing of the channel estimation over uplink signals, which comprises a delay spread. The system can determine a number of symbols for a second common PUCCH format 1 communication format with the user equipment based on the first channel characteristic. The system can determine a group of initial cyclic shift indices for the second common PUCCH format 1 communication format with the user equipment based on the second channel characteristic. The system can communicate with the user equipment according to the second common PUCCH format 1 communication format.

Abstract: A computer-implemented system for encoding includes an encoding layer and at least one joint encoding unit. The encoding layer encodes a received first modal initial feature vector and a received second modal initial feature vector, to generate, respectively, a first modal feature vector and a second modal feature vector, joint encoded by the at least one joint encoding unit, where the at least one joint encoding unit includes an encoding module and a modal input switching module. The modal input switching module processes the first modal feature vector and the second modal feature vector, to obtain, respectively a first modal switching encoding vector and a second modal switching encoding vector. The encoding module processes the first modal switching encoding vector and the second modal switching encoding vector, to generate, respectively a first target modal fusion vector and a second target modal fusion vector.

Abstract: The present invention relates to a joint automatic audio visual driven facial animation system that in some example embodiments includes a full scale state of the art Large Vocabulary Continuous Speech Recognition (LVCSR) with a strong language model for speech recognition and obtained phoneme alignment from the word lattice.

Abstract: The present invention relates to a joint automatic audio visual driven facial animation system that in some example embodiments includes a full scale state of the art Large Vocabulary Continuous Speech Recognition (LVCSR) with a strong language model for speech recognition and obtained phoneme alignment from the word lattice.

Abstract: A knowledge graph processing method is provided. The method includes: selecting several nodes and their edges from a shared knowledge graph based on one or more entity types involved in a target service domain, to obtain a target subgraph, where the shared knowledge graph is obtained by fusing knowledge graphs of one or more service domains; processing the target subgraph to extract one or more graph features, where the graph feature includes some or all of the following: a node representation vector, an edge representation vector, a graph structure feature, a semantic feature of graph text information, and a graph rule feature; and providing the graph feature to a target data processing task of the target service domain, where the graph feature is used to serve as an input feature of the target data processing task together with a task customization feature.

Abstract: Implementations of the present specification provide a digital avatar recommendation method and recommendation system. The digital avatar recommendation system includes a computer-simulated digital avatar, and the corresponding recommendation method includes: obtaining current state data, where the state data includes user information of a target user, scenario information of a current scenario, and history information of an interaction between the target user and the digital avatar; mapping, by an agent in the digital avatar, the state data to a target action in a candidate action set based on a current policy obtained through reinforcement learning, where a candidate action in the candidate action set corresponds to a to-be-recommended content category, and the target action corresponds to a target content category; and performing, by the digital avatar, target interaction with the target user, where the target interaction is used to recommend the target content category.

Abstract: A computer-implemented system for encoding includes an encoding layer and at least one joint encoding unit. The encoding layer encodes a received first modal initial feature vector and a received second modal initial feature vector, to generate, respectively, a first modal feature vector and a second modal feature vector, joint encoded by the at least one joint encoding unit, where the at least one joint encoding unit includes an encoding module and a modal input switching module. The modal input switching module processes the first modal feature vector and the second modal feature vector, to obtain, respectively a first modal switching encoding vector and a second modal switching encoding vector. The encoding module processes the first modal switching encoding vector and the second modal switching encoding vector, to generate, respectively a first target modal fusion vector and a second target modal fusion vector.

Abstract: Implementations of the present specification disclose methods, apparatuses, and devices for training an object detection system by using a gradient fine-tuning technique. In one aspect, the method includes: providing a training image as input to the object detection system; processing the training image by the object detection system; determining a gradient norm of each neural network layer based on object annotation data and the detection result corresponding to the training image; and updating, for each neural network layer, parameter values of the neural network layer based on an average of gradient norms and the gradient norm of the neural network layer.

Abstract: Implementations of the present specification disclose a knowledge graph reasoning method and apparatus, a model training method and apparatus, and a computer device. The method includes: obtaining a query entity and a query relationship; selecting one or more nearest neighbor entities of the query entity from a knowledge graph; determining a first probability of a nearest neighbor entity of the one or more nearest neighbor entities, where the first probability is used to indicate a possibility that the nearest neighbor entity is in communication with the query relationship; selecting a nearest neighbor entity of the one or more nearest neighbor entities as a candidate entity based on the first probability; and selecting a candidate entity matching the query entity and the query relationship as a result entity. In the implementations of the present specification, the efficiency of knowledge graph reasoning can be improved.

Abstract: The present disclosure provides multi-sampling model training methods and devices. One exemplary training method includes: performing multi-sampling on samples to obtain a training set and a validation set in each sampling; using the training set and the validation set obtained in each sampling as a group, and performing model training and obtaining a trained model using the training set in each group; evaluating the trained model using the training set and the validation set in each group separately; eliminating or retaining the trained model based on the evaluation results and a predetermined elimination criterion; obtaining prediction results of the samples using retained models; and obtaining a final model by performing combined model training on the retained models using the prediction results. The final model obtained using embodiments of the present disclosure can be more robust and stable, and can provide more accurate prediction results, thus greatly improving efficiency of modeling.

Abstract: The present invention relates to a joint automatic audio visual driven facial animation system that in some example embodiments includes a full scale state of the art Large Vocabulary Continuous Speech Recognition (LVCSR) with a strong language model for speech recognition and obtained phoneme alignment from the word lattice.

Abstract: Symbol prediction can be implemented using a multi-task system trained for different tasks. The tasks may include a single symbol prediction, symbol category prediction, and symbol subcategory prediction. Categories of symbols can be generated by clustering sets of training data using a clustering scheme.

Abstract: Embodiments of the present disclosure provide a method and system for training a model by using training data. The training data includes a plurality of samples, each sample includes N features, and features in the plurality of samples form N feature columns, and the method includes: determining an importance value of each of the N feature columns; determining whether the importance value of each of the N feature columns satisfies a threshold condition; performing a dimension reduction on M feature columns to generate P feature columns in response to the determination that the importance values of the M feature columns do not satisfy the threshold condition, wherein M<N and P<M; merging (N?M) feature columns having importance values that satisfy the threshold condition and the generated P feature columns to obtain (N?M+P) feature columns; and training the model based on the training data including the (N?M+P) feature columns.

Abstract: The present invention relates to a joint automatic audio visual driven facial animation system that in some example embodiments includes a full scale state of the art Large Vocabulary Continuous Speech Recognition (LVCSR) with a strong language model for speech recognition and obtained phoneme alignment from the word lattice.

Abstract: Software for online active learning receives content posted to an online stream at a website. The software converts the content into an elemental representation and inputs the elemental representation into a probit model to obtain a predictive probability that the content is abusive. The software also calculates an importance weight based on the elemental representation. And the software updates the probit model using the content, the importance weight, and an acquired label if a condition is met. The condition depends on an instrumental distribution. The software removes the content from the online stream if a condition is met. The condition depends on the predictive probability, if an acquired label is unavailable.

Abstract: The present disclosure provides methods and an apparatuses for building a data identification model. One exemplary method for building a data identification model includes: performing logistic regression training using training samples to obtain a first model, the training samples comprising positive and negative samples; sampling the training samples proportionally to obtain a first training sample set; identifying the positive samples using the first model, and selecting a second training sample set from positive samples that have identification results after being identified using the first model; and performing Deep Neural Networks (DNN) training using the first training sample set and the second training sample set to obtain a final data identification model. The methods and the apparatuses of the present disclosure improve the stability of data identification models.

Abstract: During operation, the system receives a customer request. Next, the system segments the customer request into customer request sentences. The system then encodes each sentence from the customer request with information sequentially collected from the previously observed sentences. Next, the system translates the encodings to sparse probabilities that measure the importance of sentences from the customer request. The system then extracts relevant sentences from the customer request based on the importance. Next, the system forms an extracted-sentence customer request embedding from embeddings for the extracted relevant customer request sentences. The system then uses the extracted-sentence customer request embedding to select an agent response from a set of possible agent responses based on comparisons between the extracted-sentence customer request embedding and embeddings for the set of possible agent responses.

Abstract: An English pronunciation assessment method includes: receiving an audio file including an English speech and a text transcript corresponding to the English speech; inputting audio signal to one or more acoustic models to obtain phonetic information of each phone in each word, wherein the one or more acoustic models are trained with speeches spoken by native speakers and further with speeches spoken by non-native speakers without labeling out mispronunciations, such that a pronunciation error is detected more accurately based on the obtained phonetic information; extracting time series features of each word; inputting the extracted time series features of each word, the obtained phonetic information of each phone in each word, and the audio signal included in the audio file to a lexical stress model to obtain misplaced lexical stress in each of words in the English speech with different number of syllables without expanding short words to cause input approximation.

Abstract: The present disclosure provides feature data processing methods and devices. One exemplary feature data processing method comprises: classifying features into an important feature set and an auxiliary feature set according to information attribute values of the features; converting features in the auxiliary feature set to hash features; and combining the hash features with features in the important feature set, and setting the combined features as fingerprint features. Training and prediction of to-be-processed data can be performed based on the fingerprint features. With the embodiments of the present disclosure, training dimensions can be more controllable and the amount training data amount can be reduced. Therefore, the efficiency of data processing can be improved.