An exemplary model search may provide optimal explainable models based on a dataset. An exemplary embodiment may identify features from a training dataset, and may map feature costs to the identified features. The search space may be sampled to generate initial or seed candidates, which may be chosen based on one or more objectives and/or constraints. The candidates may be iteratively optimized until an exit condition is met. The optimization may be performed by an external optimizer. The external optimizer may iteratively apply constraints to the candidates to quantify a fitness level of each of the seed candidates. The fitness level may be based on the constraints and objectives. The candidates may be a set of data, or may be trained to form explainable models. The external optimizer may optimize the explainable models until the exit conditions are met.

One embodiment provides a system which facilitates construction of an ensemble of machine learning models. During operation, the system determines a training set of data objects, wherein each data object is associated with one of a plurality of classes. The system divides the training set of data objects into a number of partitions. The system generates a respective machine learning model for each respective partition using a universal kernel function, which processes the data objects divided into a respective partition to obtain the ensemble of machine learning models. The system trains the machine learning models based on the data objects of the training set. The system predicts an outcome for a testing data object based on the ensemble of machine learning models and an ensemble decision rule.

A data generating method includes obtaining first sample data, determining a type of the first sample data and a corresponding data expansion method, expanding the first sample data according to the determined data expansion method to generate second sample data, and dividing the first sample data and the second sample data into a training set and a verification set according to a preset rule. A data model is trained according to the training set, and the data model is verified according to the verification set after training.

An example system includes a processor to evaluate a trained first classifier on a test set of labeled data to generate error rates for a number of labels. The processor is to process a set of unlabeled data via the trained first classifier to generate annotated data including labels and associated error rates. The processor is to train a second classifier using the annotated data and the associated error rates.

Provided are a method and apparatus for evaluating a joint training model. A specific implementation of the method for evaluating a joint training model comprises: receiving a model evaluation data request sent by a target device, wherein the target device comprises a participant of a joint training model; acquiring a sample set matching the model evaluation data request, wherein the matching sample set is labeled data associated with the joint training model; and generating model evaluation data of the joint training model according to the matching sample set. By means of the implementation, an effect index of a joint training model can be shared on the premise of not exposing original sample data. Accordingly, a timely and effective data reference basis is provided for the optimization and improvement of the joint training model.

Systems and methods for scheduling environment perception-based data offloading for numerous connected vehicles are disclosed. In one embodiment, a method for offloading data includes capturing an image of a view of interest from a vehicle, segmenting the image into a plurality of blocks, and determining a scheduling priority for each of one or more blocks among the plurality of blocks based on block values, wherein the block values relate to one or more objects of interest contained in each of the one or more blocks. The method further includes offloading, from the vehicle to a server, one or more blocks based on the scheduling priority of the one or more blocks.

Embodiments are directed to pre-training a transformer model using more parameters for sophisticated patterns (PSP++). The transformer model is divided into a held-out model and a main model. A forward pass and a backward pass are performed on the held-out model, where the forward pass determines self-attention hidden states of the held-out model and the backward pass determines loss of the held-out model. A forward pass on the main model is performed to determine a self-attention hidden states of the main model. The self-attention hidden states of the main model are concatenated with the self-attention hidden states of the held-out model. A backward pass is performed on the main model to determine a loss of the main model. The parameters of the held-out model are updated to reflect the loss of the held-out model and parameters of the main model are updated to reflect the loss of the main model.

An electronic device includes a graphic processor and a memory device. The graphic processor includes an artificial neural network engine that makes an object recognition model learn by using learning data and weights to provide a learned object recognition model. The memory device divides a feature vector into a first sub feature vector and a second feature vector, and performs a first calculation to apply the second sub feature vector and the weights to the learned object recognition model to provide a second object recognition result. The artificial neural network engine performs a second calculation to apply the first sub feature vector and the weights to the learned object recognition model to provide a first object recognition result and provides the first object recognition result to the memory device. The second calculation is performed in parallel with the first calculation.

There are provided a method and a system for image segmentation utilizing a GAN architecture. A method for training an image segmentation network according to an embodiment includes: inputting an image to a first network which is trained to output a region segmentation result regarding an input image, and generating a region segmentation result; and inputting the region segmentation result generated at the generation step and a ground truth (GT) to a second network, and acquiring a discrimination result, the second network being trained to discriminate inputted region segmentation results as a result generated by the first network and a GT, respectively; and training the first network and the second network by using the discrimination result. Accordingly, region segmentation performance of a semantic segmentation network regarding various images can be enhanced, and a very small image region can be exactly segmented.

Embodiments provide systems, methods, and non-transitory computer storage media for providing search result images based on associations of keywords and depth-levels of an image. In embodiments, depth-levels of an image are identified using depth-map information of the image to identify depth-segments of the image. The depth-segments are analyzed to determine keywords associated with each depth-segment based on objects, features, or content in each depth-segment. An image depth-level data structure is generated by matching keywords generated for the entire image with the keywords at each depth-level and assigning the depth-level to the keyword in the image depth-level data structure for the entire image. The image depth-level data structure may be queried for images that contain keywords and depth-level information that match the keywords and depth-level information specified in a search query.