A neural network architecture and method for analysis of time series images from an image source employs a 3D-UNet convolutional neural network (CNN) configured to receive the time series images and generate spatiotemporal feature maps therefrom. Multiple sub-convolutional neural network output prongs based on an SRNet architecture receive the feature maps and simultaneously generate inferences for image segmentation, regression of values, and multi-landmark localization.

A well log is scanned for one or more dimensions that describe one or more features of a well. Each dimension includes a plurality of values in a numerical format that represents each dimension. A missing value is detected in a first plurality of values of a first dimension of the well log. The first dimension of the well log is transformed, in response to the missing value, into a first image that visually depicts the first dimension including the first plurality of values and the missing value. Based on the first image and based on an image analysis algorithm a second image is created that visually depicts the first plurality of values and includes a found depiction visually depicting a found value in place of the missing value. The found depiction is converted, based on the second image, into a first value in the numerical format.

Embodiments relate to a system, program product, and method for automatically determining which activation data points in a neural model have been poisoned to erroneously indicate association with a particular label or labels. A neural network is trained using potentially poisoned training data. Each of the training data points is classified using the network to retain the activations of the last hidden layer, and segment those activations by the label of corresponding training data. Clustering is applied to the retained activations of each segment, and a cluster assessment is conducted for each cluster associated with each label to distinguish clusters with potentially poisoned activations from clusters populated with legitimate activations. The assessment includes executing a set of analyses and integrating the results of the analyses into a determination as to whether a training data set is poisonous based on determining if resultant activation clusters are poisoned.

Embodiments relate to a system, program product, and method for automatically determining which activation data points in a neural model have been poisoned to erroneously indicate association with a particular label or labels. A neural network is trained using potentially poisoned training data. Each of the training data points is classified using the network to retain the activations of the last hidden layer, and segment those activations by the label of corresponding training data. Clustering is applied to the retained activations of each segment, and a cluster assessment is conducted for each cluster associated with each label to distinguish clusters with potentially poisoned activations from clusters populated with legitimate activations. The assessment includes executing a set of analyses and integrating the results of the analyses into a determination as to whether a training data set is poisonous based on determining if resultant activation clusters are poisoned.

Methods, systems, apparatus, and computer-readable media for data band selection using machine learning. In some implementations, image data comprising information for each of multiple wavelength bands is obtained. A multi-layer neural network is trained using the image data to perform one or more classification or regression tasks. A proper subset of the wavelength bands is selected based on parameters of a layer of the trained multi-layer neural network, where the parameters were determined through training of the multi-layer neural network using the image data. Output is provided indicating that the selected wavelength bands are selected for the one or more classification or regression tasks.

A method, apparatus, and non-transitory computer readable medium for image processing are described. Embodiments of the method, apparatus, and non-transitory computer readable medium include identifying an original image including a plurality of semantic attributes, wherein each of the semantic attributes represents a complex set of features of the original image; identifying a target attribute value that indicates a change to a target attribute of the semantic attributes; computing a modified feature vector based on the target attribute value, wherein the modified feature vector incorporates the change to the target attribute while holding at least one preserved attribute of the semantic attributes substantially unchanged; and generating a modified image based on the modified feature vector, wherein the modified image includes the change to the target attribute and retains the at least one preserved attribute from the original image.

Provided are a learning data collection apparatus, a learning data collection method, and a program for collecting learning data to be used for efficient retraining. A learning data collection apparatus (10) includes an inspection image acquisition unit (11) that acquires an inspection image, a region detection result acquisition unit (damage detection result acquisition unit (13)) that acquires a region detection result the region detection result indicating a region detected by a region detector that is trained, a correction history acquisition unit (15) that acquires a correction history of the region detection result, a calculation unit (17) that calculates correction quantification information obtained by quantifying the correction history, a database that stores the inspection image, the region detection result, and the correction history in association with each other, an image extraction condition setting unit (19) that sets a threshold value of the correction quantification information as an extraction condition, the extraction condition being a condition for extracting the inspection image to be used for retraining from the database, and a first learning data extraction unit (21) that extracts, as learning data for retraining the region detector, the inspection image satisfying the extraction condition and the region detection result and the correction history that are associated with the inspection image.

Provided are a learning data collection apparatus, a learning data collection method, and a program for collecting learning data to be used for efficient retraining. A learning data collection apparatus (10) includes an inspection image acquisition unit (11) that acquires an inspection image, a region detection result acquisition unit (damage detection result acquisition unit (13)) that acquires a region detection result the region detection result indicating a region detected by a region detector that is trained, a correction history acquisition unit (15) that acquires a correction history of the region detection result, a calculation unit (17) that calculates correction quantification information obtained by quantifying the correction history, a database that stores the inspection image, the region detection result, and the correction history in association with each other, an image extraction condition setting unit (19) that sets a threshold value of the correction quantification information as an extraction condition, the extraction condition being a condition for extracting the inspection image to be used for retraining from the database, and a first learning data extraction unit (21) that extracts, as learning data for retraining the region detector, the inspection image satisfying the extraction condition and the region detection result and the correction history that are associated with the inspection image.

In a model generation apparatus, a training data selection means selects sets of training data from a plurality of fingerprint images. A learning means trains a model that corrects a fingerprint image, by using the sets of training data. An evaluation value calculation means calculates evaluation values of results acquired by inputting the plurality of fingerprint images into the trained model. A model update means updates a model to be trained, based on the evaluation values. Next, the training data selection means determines sets of training data to be selected from the plurality of fingerprint images based on the evaluation values.

In a model generation apparatus, a training data selection means selects sets of training data from a plurality of fingerprint images. A learning means trains a model that corrects a fingerprint image, by using the sets of training data. An evaluation value calculation means calculates evaluation values of results acquired by inputting the plurality of fingerprint images into the trained model. A model update means updates a model to be trained, based on the evaluation values. Next, the training data selection means determines sets of training data to be selected from the plurality of fingerprint images based on the evaluation values.