A content aware and style aware neural network based data augmentation model generates augmented data sets to train neural network based handwriting recognition models to recognize individuals' handwriting. In embodiments, the augmented data sets are generated so as to be artificial, and to lack personal or confidential information. In embodiments, the data augmentation model generates content reference sets of individual characters generated in different fonts, and style reference sets of pluralities of characters of a particular style, for example, an individual's handwriting.

A computer-implement method comprises: training a classifier with labeled data from a dataset; classifying, by the trained classifier, unlabeled data from the dataset; providing, by the classifier to a policy gradient, a reward signal for each data/query pair; transferring, by the classifier to a ranker, learning; training, by the policy gradient, the ranker; ranking data from the dataset based on a query; and retrieving data from the ranked data in response to the query.

Devices and techniques are generally described for cross-domain machine learning. A first machine learning model may be trained using first data of a first domain. Predictions may be generated by inputting a plurality of domain data from other domains apart from the first domain into the first machine learning model. For each of the predictions, a prediction error may be determined. A grouping of similar domains from among the other domains may be determined based on the prediction errors. A second machine learning model may be trained for the grouping of similar domains.

A method for training a course recommendation model, a method for course recommendation, and an apparatus, which relate to a field of big data and deep learning in a field of artificial intelligence technology, and can be applied to recommendation scenarios. The training method includes: obtaining a sample data set, where the sample data set includes user learning data, the user learning data includes record data and ability label data, the record data is used for representing a historical learning process of a sample user, and the ability label data is used for representing a learning ability level of the sample user, and training and generating the course recommendation model according to the user learning data, where the course recommendation model is used for recommending a course for a user, the technical effect of improving the reliability and accuracy of course recommendation is achieved.

A memory access method used when training a deep learning network is illustrated in the present disclosure. When calculating the weightings of the current layer to the previous layer, the differential terms generated by the weighting updating calculation from the next layer to the current layer are used for reducing the access number of accessing the memory. Since the memory access method greatly reduces the access number of accessing the memory, the training time and power consumption can be reduced, and the lifetime of the battery and memory of the deep learning network device can be prolonged. Especially in the case of limited battery power, the deep learning network device can run longer.

Methods and systems for training a neural network include training language-specific teacher models using different respective source language datasets. A student model is trained, using the different respective source language datasets and soft labels generated by the language-specific teacher models, including shuffling the source language datasets and shuffling weights of language-dependent layers in language-specific parts of the student model. Weights of language-independent layers of the student model are copied to a language-independent layers of a target model to initialize language-independent layers of the target model. The target model is trained with a target language dataset.

A method of balancing a dataset for a machine learning model includes identifying confusing classes of few-shot classes for a machine learning model during validation. One of the confusing classes and an image from one of the few-shot classes are selected. An image perturbation is computed such that the selected image is classified as the selected confusing class. The selected image is modified with the computed perturbation. The modified selected image is added to a batch for training the machine learning model.

A non-transitory computer-readable recording medium stores a program that causes a computer to execute a process, the process includes inputting each of first images that includes a face of a subject to a first machine learning model to obtain a recognition result that includes information indicating first occurrence probability of each of facial expressions in each first image, generating training data that includes the recognition result and second images that are respectively generated based on the first images and in which at least a part of the face of the subject is concealed, and performing training of a second machine learning model, based on the training data, by using a loss function that represents an error that relates to a second occurrence probability of each facial expression in each second image and relates to magnitude relationship in the second occurrence probability among the second images.

The technology described herein determines whether a candidate text is in a requested class by using a generative model that may not be trained on the requested class. The present technology may use of a model trained primarily in an unsupervised mode, without requiring a large number of manual user-input examples of a label class. The may produce a semantically rich positive example of label text from a candidate text and label. Likewise, the technology may produce from the candidate text and the label a semantically rich negative example of label text. The labeling service makes use of a generative model to produce a generative result, which estimates the likelihood that the label properly applies to the candidate text. In another aspect, the technology is directed toward a method for obtaining a semantically rich example that is similar to a candidate text.

A crop yield estimation method based on spatio-temporal deep learning including: obtaining regional historical crop yield data and meteorological data, preprocessing the meteorological data and the yield data to respectively obtain meteorological parameters and a detrended yield as input and output of the crop yield spatio-temporal deep learning model; constructing the spatio-temporal deep learning model for crop yield estimation, and optimizing hyperparameters; and building a training set by taking the meteorological parameters as an input and the detrended yield as output to train the model and obtain parameters of the model; for the crop yield to be estimated, feeding meteorological parameters into the trained model, and obtaining the crop yield estimation result. The model combined temporal and spatial learning to achieve better crop yield estimation accuracy and stability at large spatial scales.