Techniques are described for domain adaptation of image processing models using post-processing model correction According to an embodiment, a method comprises training, by a system operatively coupled to a processor, a post-processing model to correct an image-based inference output of a source image processing model that results from application of the source image processing model to a target image from a target domain that differs from a source domain, wherein the source image processing model was trained on source images from the source domain. In one or more implementations, the source imaging processing model comprises an organ segmentation model and the post-processing model can comprise a shape-autoencoder. The method further comprises applying, by the system, the source image processing model and the post-processing model to target images from the target domain to generate optimized image-based inference outputs for the target images.

A method is disclosed for evaluating a classifier used to determine a traffic light signal state in images. The method includes, by a computer vision system of a vehicle, receiving at least one image of a traffic signal device of an imminent intersection. The traffic signal device includes a traffic signal face including one or more traffic signal elements. The method includes classifying, by a traffic light classifier (TLC), a classification state of the traffic signal face using labeled images correlated to the received at least one image. The classification state controls an operation of the vehicle at the intersection. The method includes evaluating a performance of the classifying of the classification state generated by the TLC. The evaluation is a label-free performance evaluation based on unlabeled images. The method includes training the TLC based on the evaluated performance.

A deep learning system establishes a simple process of generating a deep learning model, and provides an intuitive, natural and easy interaction in performing feedback on image input, manual labelling and automated labelling required for the above-described operations. Therefore, a user without expertise in deep learning can have an opportunity to directly generate and use a user-customized image identification deep learning model for identifying a desired object to be identified.

Embodiments relate to a system, program product, and method for information extraction and annotation of a data set. Neural models are utilized to automatically attach machine annotations to data elements within an unlabeled data set. The attached machine annotations are evaluated and a score is attached to the annotations. The score reflects a confidence of correctness of the annotations. A labeled data set is iteratively expanded with selectively evaluated annotations based on the attached score. The labeled data set is applied to an unexplored corpus to identify matching corpus data to populated instances of the labeled data set.

The present disclosure relates to systems, devices, and methods to reconstruct a three-dimensional model of an anatomy using two-dimensional images.

The disclosed technology provides solutions for finding samples from image data that are similar to failure cases, by constructing N-dimensional vectors of the failure cases. The vectors of failure cases are compared to other image data, with the objective of identifying groups of images that can be labeled. The labeled images are then used to retrain a model. Systems and machine-readable media are also provided.

In one embodiment, a computing system of a vehicle generates perception data based on sensor data captured by one or more sensors of the vehicle. The perception data includes one or more representations of physical objects in an environment associated with the vehicle. The computing system further determines simulated perception data that includes one or more representations of virtual objects within the environment and generates modified perception data based on the perception data and the simulated perception data. The modified perception data includes at least one of the one or more representations of physical objects and the one or more representations of virtual objects. The computing system further determines a path of travel for the vehicle based on the modified perception data, which includes the one or more representations of the virtual objects.

A fault diagnosis method for power electronic circuits based on optimizing a deep belief network, including steps. (1) Use RT-LAB hardware-in-the-loop simulator to set up fault experiments and collect DC-link output voltage signals in different fault types. (2) Use empirical mode decomposition to extract the intrinsic function components of the output voltage signal and its envelope spectrum and calculate various statistical features to construct the original fault feature data set. (3) Based on the feature selection method of extreme learning machine, remove the redundancy and interference features, as fault sensitive feature data set. (4) Divide the fault sensitive feature set into training samples and test samples, and primitively determine the structure of the deep belief network. (5) Use the crow search algorithm to optimize the deep belief network. (6) Obtain the fault diagnosis result.

The current document is directed to methods and systems that effectively and efficiently employ incomplete training data to train machine-learning-based systems. Incomplete training data, as one example, may include training data with erroneous or inaccurate input-vector/label pairs. In currently disclosed methods and systems, Incomplete training data is mapped to loss classes based on addition training-data information and specific, different additional-information-dependent loss-generation methods are employed for training data of different loss classes during machine-learning-based-system training so that incomplete training data can be effectively and efficiently used.

Embodiments relate to a system, program product, and method for processing an untrusted data set to automatically determine which data points there are poisonous. A neural network is trained network using potentially poisoned training data. Each of the training data points is classified using the network to retain the activations of at least one 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 clustering assessment is conducted to remove an identified cluster from the data set, form a new training set, and train a second neural model with the new training set. The removed cluster and corresponding data are applied to the trained second neural model to analyze and classify data in the removed cluster as either legitimate or poisonous.