A system includes a first neural network configured to calculate, based on input data, data indicative of a predicted result of a predetermined prediction task for the input data, and a second neural network configured to calculate, based on the input data and labelled data corresponding to the input data, data related to error in the labelled data. At least one of the first neural network or the second neural network is trained by using at least both the data indicative of the predicted result calculated by the first neural network and the data related to the error in the labelled data calculated by the second neural network.

An on-road obstacle detection device that includes: a memory; and a processor, the processor being connected to the memory and being configured to: assign a semantic label to each pixel in an image using a first discriminator that has been pre-trained using images in which an on-road obstacle is not present; and detect an on-road obstacle based on a probability density of the semantic label assigned.

A method for training an image classification model according to an embodiment includes training a feature extractor and a rotation angle classifier to predict a rotation angle of each of unlabeled first training images, training the image classification model to predict a label and rotation angle of each of labeled second training images, but predict a uniform label even though an actual rotation angle of each of the second training images is changed, generating a pseudo label based on a training image that satisfy a preset condition among unlabeled candidate images, and training the image classification model to predict a rotation angle of each of the third training images, and predict a label of each of the third training images based on the pseudo label, but predict a uniform label even though an actual rotation angle of each of the third training images is changed.

In a method for defecting surface defects, a trained weighting generated when defect-free training samples are used to train an autoencoder and pixel convolutional neural network is obtained. A test encoding feature is obtained by inputting the trained weighting into the autoencoder and pixel convolutional neural network and a weighted autoencoder of the weighted autoencoder and pixel convolutional neural network encoding a test sample. The test encoding feature is input into a weighted pixel convolution neural network of the weighted autoencoder and pixel convolutional neural network to output a result of test. The test result is either no defect in the test sample or at least one defect in the test sample. Inaccurate determinations as to defects are thereby avoided. An electronic device and a non-transitory storage medium are also disclosed.

A method of training a cycle generative networks model and a method of building a character library are provided, which relate to a field of artificial intelligence, in particular to a computer vision and deep learning technology, and which may be applied to a scene such as image processing and image recognition. A specific implementation scheme includes: inputting a source domain sample character into the cycle generative networks model to obtain a first target domain generated character; calculating a character error loss and a feature loss of the cycle generative networks model by inputting the first target domain generated character and a preset target domain sample character into a character classification model; and adjusting a parameter of the cycle generative networks model according to the character error loss and the feature loss. An electronic device and a storage medium are further provided.

A classifier is executed including an unsupervised artificial intelligence model and a supervised artificial intelligence model. The classifier is configured to receive run-time input data, and process the run-time input data using the unsupervised artificial intelligence model and an outlier detection algorithm to determine whether the run-time input data is an outlier as compared to training input data. Responsive to determining that the run-time input data is not an outlier, the classifier determines a predicted response label for the run-time input based on the run-time input data processed using the supervised artificial intelligence model. Responsive to determining that the run-time input data is an outlier, the classifier refrains from determining the predicted response label for the run-time input based on the run-time input data processed using the supervised artificial intelligence model, and instead outputs a prompt for user input of a user-curated response label for the run-time input.

The subject disclosure presents systems and computer-implemented methods for providing reliable risk stratification for early-stage cancer patients by predicting a recurrence risk of the patient and to categorize the patient into a high or low risk group. A series of slides depicting serial sections of cancerous tissue are automatically analyzed by a digital pathology system, a score for the sections is calculated, and a Cox proportional hazards regression model is used to stratify the patient into a low or high risk group. The Cox proportional hazards regression model may be used to determine a whole-slide scoring algorithm based on training data comprising survival data for a plurality of patients and their respective tissue sections. The coefficients may differ based on different types of image analysis operations applied to either whole-tumor regions or specified regions within a slide.

Some embodiments of the current disclosure disclose methods and systems for batch active learning using the Shapley values of data points. In some embodiments, Shapley values of a first subset of labeled data are used to measure the contributions of the first subset of data to the performance of neural network. Further, a regression model that correlates the first subset of data to their Shapley values is trained to predict the Shapley values of a second subset of data that are unlabeled. A portion of the second subset of data may then be selected for labeling based on the predicted Shapley values.

A combination of two or more trained machine learning models can exhibit a combined accuracy greater than the accuracy of any one of the constituent models. However, this increase accuracy comes at additional computational cost. Cascades of machine learning models are provided herein that result in increased model accuracy and/or reduced model compute cost. These benefits are obtained by conditionally executing one or more of the models of the cascade based on the estimated correctness of already-executed models. The estimated correctness can be obtained as an additional output of the already-executed model(s) or could be determined as an entropy, maximum class probability, maximum class logit, or other function of the output(s) of the already-executed model(s). The expected computational cost of executing the model cascade is reduced by only executing the downstream model(s) when the upstream model(s) has resulted in an output whose accuracy is suspect.

A medical system may utilize a modular and extensible sensing device to derive a two-dimensional (2D) or three-dimensional (3D) human model for a patient in real-time based on images of the patient captured by a sensor such as a digital camera. The 2D or 3D human model may be visually presented on one or more devices of the medical system and used to facilitate a healthcare service provided to the patient. In examples, the 2D or 3D human model may be used to improve the speed, accuracy and consistency of patient positioning for a medical procedure. In examples, the 2D or 3D human model may be used to enable unified analysis of the patient's medical conditions by linking different scan images of the patient through the 2D or 3D human model. In examples, the 2D or 3D human model may be used to facilitate surgical navigation, patient monitoring, process automation, and/or the like.