Robert Boschgmbh

Carnegie Mellon University

Jan Hendrik Metzen

Volker Fischer

Chaithanya Kumar Mummadi

Robin Hutmacher

Jeremy Zico Kolter

Kilian Rambach

Nicole Ying Finnie

Wan-Yi Lin

Jeremy Zieg Kolter

Frank R Schmidt

Frank Hutter

Frank Schmidt

Fatemeh Sheikholeslami

Benedikt Sebastian Staffler

Eric Wong

Huan Zhang

Jan Biel

Evgeny Levinkov

Anurag Pandey

Thomas Elsken

Solveig Klepper

A computer-implemented method for determining an output signal characterizing a first classification of an input image into a class from a plurality of classes. The output signal further characterizes a second classification of a robustness of the first classification against an attack with an adversarial patch.

Device and method for classifying images and accessing the robustness of the classification

A computer-implemented method for adapting a pretrained machine learning system, which has been trained on a first training data set, to a second dataset, wherein the second dataset has different characteristics than the first data set. An input transformation module for partly undoing the distribution shift between the first and second training data set is provided.

Device and method to adapt a pretrained machine learning system to target data that has different distribution than the training data without the necessity of human annotations on target data

A computer-implemented method for determining an output signal characterizing a semantic segmentation and/or an instance segmentation of an image. The method includes: determining a first intermediate output signal from a machine learning system, wherein the first intermediate output signal characterizes a semantic segmentation and/or an instance segmentation of the image; adapting parameters of the machine learning system based on a loss function, wherein the loss function characterizes an entropy or a cross-entropy of the first intermediate output signal; determining the output signal from the machine learning system based on the image and the adapted parameters.

Device and method for determining a semantic segmentation and/or an instance segmentation of an image

A system includes a machine-learning network. The network includes an input interface configured to receive input data from a sensor. The processor is programmed to receive the input data, generate a perturbed input data set utilize the input data, wherein the perturbed input data set includes perturbations of the input data, denoise the perturbed input data set utilizing a denoiser, wherein the denoiser is configured to generate a denoised data set, send the denoised data set to both a pre-trained classifier and a rejector, wherein the pre-trained classifier is configured to classify the denoised data set and the rejector is configured to reject a classification of the denoised data set, train, utilizing the denoised input data set, the a rejector to achieve a trained rejector, and in response to obtaining the trained rejector, output an abstain classification associated with the input data, wherein the abstain classification is ignored for classification.

System and method for improving robustness of pretrained systems in deep neural networks utilizing randomization and sample rejection

A system and method, in particular computer implemented method for determining a perturbation for attacking and/or validating an association tracker. The method includes providing digital image data that includes an object, determining with the digital image data a first feature that characterizes the object, providing in particular from a storage a second feature that characterizes a tracked object, determining the perturbation depending on a measure of a similarity between the first feature and the second feature.

System and method, in particular computer implemented method for determining a perturbation for attacking and/or validating an object tracker

A method for creating a machine learning system which is designed for segmentation and object detection in images. The method includes: providing a directed graph; selecting a path through the graph, at least one additional node being selected from this subset, a path through the graph from the input node along the edges via the additional node up to the output node being selected; creating a machine learning system as a function of the selected path; and training the machine learning system created.

Method and device for creating a machine learning system

A method for measuring the sensitivity of a classifier for digital images against adversarial attacks. The classifier includes at least one neural network. The method includes: providing a digital image for which the sensitivity is to be measured; providing a generator that is trained to map elements of a latent space to a realistic image; obtaining, according to a set of parameters, an element of the latent space; mapping, using the generator, this element to a disturbance in the space of realistic images; perturbing the digital image with this disturbance; determining, using the classifier, a classification result for the perturbed image; determining, from the classification result, the impact of the disturbance on the classification result; optimizing the set of parameters to maximize this impact; and determining, based at least in part on the maximum impact, the sensitivity of the classifier.

Measuring the sensitivity of neural network image classifiers against adversarial attacks

A method for training a neural network. The method includes providing learning input images and associated learning output data; providing auxiliary input images; generating modifications of these auxiliary input images by introducing at least one predefined change into them; supplying the modifications to the neural network; ascertaining predictions for the predefined change, using output data onto which the neural network maps the modifications; assessing deviations of the predictions from the predefined change, using an auxiliary cost function; optimizing parameters of the neural network to improve the assessment by the auxiliary cost function during further processing of auxiliary input images; supplying the learning input images to the neural network; assessing deviations of the output data, thus obtained, from the learning output data, using a main cost function; optimizing parameters of the neural network to improve the assessment by the main cost function during further processing of learning input images.

Data-based updating of the training of classifier networks

A method for automatically generating an artificial neural network that encompasses modules and connections that link those modules, successive modules and/or connections being added to a current starting network. Modules and/or connections that are to be added are selected randomly from a predefinable plurality of possible modules and connections that can be added. A plurality of possible refinements of the current starting network respectively are generated by adding to the starting network modules and/or connections that are to be added. One of the refinements from the plurality of possible refinements is then selected in order to serve as a current starting network in a subsequent execution of the method.

Method and apparatus for automatically producing an artificial neural network

A method for training an automated learning system includes processing training input with a first neural network and processing the output of the first neural network with a second neural network. The input layer of the second neural network corresponding to the output layer of the first neural network. The output layer of the second neural network corresponding to the input layer of the first neural network. An objective function is determined using the output of the second neural network and a predetermined modification magnitude. The objective function is approximated using random Cauchy projections which are propagated through the second neural network.

Method, apparatus and computer program for generating robust automatic learning systems and testing trained automatic learning systems

A method for operating a detector that is set up to check whether a data signal that is supplied to a machine learning system has been manipulated. The machine learning system is first trained in adversarial fashion using a manipulated data signal, the manipulated data signal having been ascertained by manipulation of a training data signal, and the machine learning system being trained to provide in each case the same output signal when the training data signal or the manipulated data signal is supplied to it. The detector is trained using another manipulated data signal that is produced as a function of the trained machine learning system.

Method and apparatus for improving the robustness of a machine learning system

A method is described for measuring the vulnerability of an AI module to spoofing attempts, including the classification and/or regression onto which the AI module maps the update data set is ascertained as an unperturbed result for a predefined data set in the input space E; at least one perturbation S having a dimensionality d&#x3c;D is applied to the predefined data set so that at least one perturbed data set results in the input space E; the classification and/or regression onto which the AI module maps the perturbed data set is ascertained as the perturbed result; the deviation of the perturbed result from the unperturbed result is ascertained using predefined metrics; in response to the deviation satisfying a predefined criterion, it is determined that the AI module with regard to the predefined data set is vulnerable to spoofing attempts having a dimensionality d.

Measuring the vulnerability of AI modules to spoofing attempts

A method for operating e.g. a robot. The method includes the steps: acquiring a sequence of images of a space surrounding the robot within which a potential object is situated; ascertaining, using a first machine learning system, a respective first variable characterizing the potential object as a function of each of the images; ascertaining, using a second machine learning system, at least one second variable that characterizes the potential object as a function of a plurality of the first variables characterizing the potential object; controlling the robot as a function of the second variable characterizing the potential object. A computer program, a device for carrying out the method, and a machine-readable storage element on which the computer program is stored, are also described.

Method, device, product, and computer program for operating a technical system

A method for generating a universal data signal interference for generating a manipulated data signal for deceiving a first machine learning system, which is configured to ascertain a semantic segmentation of a received one-dimensional or multidimensional data signal, The method includes a) ascertaining a training data set that includes pairs of data signals and associated desired semantic segmentations, b) generating the data signal interference, as a function of the data signals of the training data set, of the associated desired semantic segmentation, as well as of estimated semantic segmentations of the data signals acted upon by the data signal interference.

Method and device for improving the robustness against 'adversarial examples'

A method for generating a manipulated data signal for misleading a first machine learning system, which is designed to ascertain a semantic segmentation of a received one-dimensional or multi-dimensional data signal, the method having the following steps: a) ascertaining a desired semantic segmentation of the manipulated data signal; and b) generating the manipulated signal as a function of the received data signal and the ascertained desired semantic segmentation as well as an estimated semantic segmentation of the manipulated data signal.

Growing community of inventors

Boeblingen, Germany

Jan Hendrik Metzen