Abstract: A method for reconstructing training examples, with which a predefined neural network has been trained to optimize a predefined cost function. A quality function is provided, which measures for a training example to what extent it belongs to an expected domain or distribution of the training examples; a variable of a batch of training examples, with which the neural network has been trained, is provided; a gradient of the cost function ascertained according to parameters, which characterize the behavior of the neural network, is divided into a partition made up of components; from each component, a training example is reconstructed using the functional dependency of the outputs of neurons in the input layer of the neural network which receives the training examples from the parameters of these neurons and from the training examples; the reconstructions obtained are assessed using the quality function; the partition into the components is optimized.

Abstract: A method for generating a semantically modified variation of an image. In the method: the image is divided into equally sized, non-overlapping patches; the patches are converted with a patch encoding function of a transformer network into a chain of tokens in a workspace; the tokens are grouped into preservation tokens, whose information is to be preserved in the variation, and masked tokens, whose information is to be masked in the variation; the preservation tokens are converted with an encoder of the transformer network into a chain of processed tokens; the chain is supplemented through application of an insertion operator that inserts these masked tokens at positions corresponding to the positions of the masked tokens in the original chain to form a chain that represents the sought variation; the chain is converted with a decoder of the transformer network into the sought variation.

Abstract: A method for training a neural network for determining a task output with respect to a given task. The method includes: providing unlabeled and/or labelled encoder training records of measurement data; training the encoder network to map encoder training records to representations towards the goal that these representations, and/or or one or more work products derived from the representations, fulfil a self-consistency condition or correspond to ground truth; providing task training records that are labelled with ground truth; and training the association network and the task head networks towards the goal that, when a task training record is mapped to a representation using the encoder network, and the representation is mapped to a task output by the combination of the association network and the task head networks, the so-obtained task output corresponds to the ground truth with which the training record is labelled, as measured by a task loss function.

Abstract: A computer-implemented method of identifying filters for use in determining explainability of a trained neural network. The method comprises obtaining a dataset comprising the input image and an annotation of an input image, the annotation indicating at least one part of the input image which is relevant for inferring classification of the input image, determining an explanation filter set by iteratively: selecting a filter of the plurality of filters; adding the filter to the explanation filter set; computing an explanation heatmap for the input image by resizing and combining an output of each filter in the explanation filter set to obtain the explanation heatmap, the explanation heatmap having a spatial resolution of the input image; and computing a similarity metric by comparing the explanation heatmap to the annotation of the input image; until the similarity metric is greater than or equal to a similarity threshold; and outputting the explanation filter set.

Abstract: A method for measuring the sensitivity of an image classifier, which assigns an input image to one or multiple classes of a predefined classification, to modifications of the input image. The method includes: the input image is mapped by at least one predefined operator onto an intermediate image, which has a lesser information content and/or a poorer signal-to-noise ratio in comparison to the input image; at least one generator is provided, which is trained to generate realistic images, which image classifier assigns to a specific class of the predefined classification; the generator is used to generate a variation of the input image from the intermediate image.

Abstract: A computer-implemented method for determining an adversarial patch for a machine learning system. The machine learning system is configured for image analysis and determines an output signal based on an input image. The output signal is determined based on an output of an attention layer of the machine learning system. The adversarial patch is determined by optimizing the adversarial patch with respect to a loss function, wherein the loss function comprises a term that characterizes a sum of attention weights of the attention layer with respect to a position of the adversarial patch in the input image and the method comprises a step of maximizing the term.

Abstract: Systems and methods for managing visual allocation are provided herein that use models to determine states based on visual data and, based thereon, output feedback based on the determined states. Visual data is initially obtained by a visual allocation management system. The visual data includes eye image sequences of a person in a particular state, such as engaging in a task or activity. Visual features can be identified from the visual data, such that glance information including direction and duration can be calculated. The visual data, information derived therefrom, and/or other contextual data is input into the models, which correspond to states, to calculate probabilities that the particular state that the person is engaged in is one of the modeled states. Based on the state identified as having the highest probability, an optimal feedback, such as a warning or instruction, can be output to a connected devices, systems, or objects.

Abstract: A method for quantitatively rating a trained generator that generates, from an input vector from a predetermined input distribution in connection with a target segmentation map, an image whose semantic content is in line with this target segmentation map. The method includes: drawing at least one input vector from the predetermined input distribution; drawing at least one list of classes from a predetermined class distribution; determining a target segmentation map, which assigns classes from this list of classes to the pixels of the image to be generated; generating, by means of the generator, an image from the input vector and the target segmentation map; determining a semantic segmentation map of the image by means of an image classifier; determining, using a predetermined metric, the degree of matching between this semantic segmentation map and the target segmentation map; determining, using this degree of matching, the quantitative rating of the generator.

Abstract: An image classifier for classifying an input image x with respect to combinations of an object value o and an attribute value. The image classifier includes an encoder network that is configured to map the input image to a representation comprising multiple independent components; an object classification head network configured to map representation components of the input image to one or more object values; an attribute classification head network configured to map representation components of the input image to one or more attribute values; and an association unit configured to provide, to each classification head network, a linear combination of those representation components of the input image x that are relevant for the classification task of the respective classification head network. A method for training the image classifier is also provided.

Abstract: A computer-implemented method of determining an explainability mask for classification of an input image by a trained neural network. The trained neural network is configured to determine the classification and classification score of the input image by determining a latent representation of the input image at an internal layer of the trained neural network. The method includes accessing the trained neural network, obtaining the input image and the latent representation thereof and initializing a mask for indicating modifications to the latent representation. The mask is updated by iteratively adjusting values of the mask to optimize an objective function, comprising i) a modification component indicating a degree of modifications indicated by the mask, and ii) a classification score component, determined by applying the indicated modifications to the latent representation and determining the classification score thereof. The mask is scaled to a spatial resolution of the input image and output.

Abstract: A method for measuring the components of an input image on which an image classifier bases its decision regarding the assignment of this input image to one or multiple class(es) of a predefined classification. The method includes: providing binary masks, which indicate which pixels of the input image and/or of an intermediate product formed in the image classifier are considered relevant; assessing the binary masks using a quality function, which is a measure of the extent to which at least one classification score, supplied by the image classifier, with respect to at least one target class changes when the pixels of the input image or of the intermediate product which are relevant according to the binary mask are changed; and ascertaining the sought-after components of the input image relevant for the decision of the image classifier from the combination of the binary masks with respective assessments by the quality function.

Abstract: A method for measuring components of an input image on which an image classifier bases its decision regarding the assignment of the input image to class(es) of a predefined classification. The method includes: processing the input image by the image classifier into an intermediate product; mapping the intermediate product on a classification score with respect to at least one target class; ascertaining a perturbation from counter image(s) which is/are preferentially assigned by the image classifier to at least one class other than the target class; providing at least one binary mask; creating at least one modification, in which pixels established by the binary mask are replaced with pixels of the perturbation corresponding thereto; mapping the modification on a classification score with respect to a predefined class; and ascertaining from the classification score to what extent the binary mask indicates the sought-after decision-relevant components of the input image.

Abstract: A system for training a generative model and a discriminative model. The generative model generates synthetic instances from latent feature vectors by generating an intermediate representation from the latent feature vector and generating the synthetic instance from the intermediate representation. The discriminative model determines multiple discriminator scores for multiple parts of an input instance, indicating whether the part is from a synthetic instance or an actual instance. The generative model is trained by backpropagation. During the backpropagation, partial derivatives of the loss with respect to entries of the intermediate representation are updated based on a discriminator score for a part of the synthetic instance, wherein the part of the synthetic instance is generated based at least in part on the entry of the intermediate representation, and wherein the partial derivative is decreased in value if the discriminator score indicates an actual instance.

Abstract: A method for measuring the sensitivity of a classifier. The method includes: at least one input data set, including image data, is provided; an initial class assignment is ascertained for the at least one input data set with the aid of the classifier; a predefined number of different faults is ascertained from the input data set based on the objectives, in that a modification of the input data set formed by the joint application of all of these faults is mapped by the classifier to a class assignment, which differs from the initial class assignment according to the stipulation of a predefined criterion; while a modification of the input data set formed by applying only one of these faults is mapped by the classifier to a class assignment, which corresponds to the initial class assignment according to the stipulation of the predefined criterion.

Abstract: Alteration of an input image to provide an interpretability of a trained classifier. The altered image may be determined by optimizing at least: a classification score of the altered image to reduce a distance between a classification of the altered image and a target classification, and a similarity score to reduce a distance between an average filter activation difference and a measure for a difference in filter activations between the input image and the altered image.

Abstract: A method for plausibilizing the output of an image classifier which assigns an input image to one or more class(es) of a predefined classification. The method includes: an assignment to one or more class(es) is ascertained for the input image using the image classifier; a relevance assessment function is used to ascertain a spatially resolved relevance assessment of the input image, which indicates which components of the input image have contributed to what degree to the assignment; a generator is trained to generate modifications of the input image that are as satisfactory as possible according to a predefined cost function in view of the optimization goals; based on the result of the training, and/or based on the modifications supplied by the trained generator, a quality measure for the spatially resolved relevance assessment, and/or a quality measure for the relevance assessment function is/are ascertained.

Abstract: A method for measuring the sensitivity of an image classifier, which assigns an input image to one or multiple classes of a predefined classification, to modifications of the input image. The method includes: the input image is mapped by at least one predefined operator onto an intermediate image, which has a lesser information content and/or a poorer signal-to-noise ratio in comparison to the input image; at least one generator is provided, which is trained to generate realistic images, which image classifier assigns to a specific class of the predefined classification; the generator is used to generate a variation of the input image from the intermediate image.

Abstract: A computer-implemented method for explaining a classification of one or more classifier inputs by a trained classifier. A generative model is used that generates inputs for the trained classifier. The generative model comprises multiple filters. Generator inputs corresponding to the one or more classifier inputs are obtained, where a generator input causes the generative model to approximately generate the corresponding classifier input. Filter suppression factors are determined for the multiple filters of the generative model. A filter suppression factor for a filter indicates a degree of suppression for a filter output of the filter. The filter suppression factors are determined based on an effect of adapting the classifier inputs according to the filter suppression factors on the classification by the trained classifier. The classification explanation is based on the filter suppression factors.

Abstract: A computer-implemented method of identifying filters for use in determining explainability of a trained neural network. The method comprises obtaining a dataset comprising the input image and an annotation of an input image, the annotation indicating at least one part of the input image which is relevant for inferring classification of the input image, determining an explanation filter set by iteratively: selecting a filter of the plurality of filters; adding the filter to the explanation filter set; computing an explanation heatmap for the input image by resizing and combining an output of each filter in the explanation filter set to obtain the explanation heatmap, the explanation heatmap having a spatial resolution of the input image; and computing a similarity metric by comparing the explanation heatmap to the annotation of the input image; until the similarity metric is greater than or equal to a similarity threshold; and outputting the explanation filter set.

Abstract: A computer-implemented method of determining an explainability mask for classification of an input image by a trained neural network. The trained neural network is configured to determine the classification and classification score of the input image by determining a latent representation of the input image at an internal layer of the trained neural network. The method includes accessing the trained neural network, obtaining the input image and the latent representation thereof and initializing a mask for indicating modifications to the latent representation. The mask is updated by iteratively adjusting values of the mask to optimize an objective function, comprising i) a modification component indicating a degree of modifications indicated by the mask, and ii) a classification score component, determined by applying the indicated modifications to the latent representation and determining the classification score thereof. The mask is scaled to a spatial resolution of the input image and output.