A system and a method generate a neural network that includes at least one layer having weights and output feature maps that have been jointly pruned and quantized. The weights of the layer are pruned using an analytic threshold function. Each weight remaining after pruning is quantized based on a weighted average of a quantization and dequantization of the weight for all quantization levels to form quantized weights for the layer. Output feature maps of the layer are generated based on the quantized weights of the layer. Each output feature map of the layer is quantized based on a weighted average of a quantization and dequantization of the output feature map for all quantization levels. Parameters of the analytic threshold function, the weighted average of all quantization levels of the weights and the weighted average of each output feature map of the layer are updated using a cost function.

Systems, methods, devices, instructions, and other examples are described for natural language processing. One example includes accessing natural language processing general encoder data, where the encoder data is generated from a general-domain dataset that is not domain specific. A domain specific dataset is accessed and filtered encoder data using a subset of the encoder data is generated. The filtered encoder data is trained using the domain specific dataset to generate distilled encoder data, and tuning values for the distilled encoder data are generated to configure task outputs associated with the domain specific dataset.

A method of generating an image recognition model for recognising an input image and a system thereof are provided. The method includes appending at least one feature extraction layer to the image recognition model, extracting a plurality of feature vectors from a set of predetermined images, grouping the plurality of feature vectors into a plurality of categories, clustering the plurality of feature vectors of each of the plurality of categories into at least one cluster, determining at least one centroid for each of the at least one cluster, such that each of the at least one cluster comprises at least one centroid, such that each of the at least one centroid is represented by a feature vector, generating a classification layer based on the feature vector of the at least one centroid of the plurality of categories, and appending the classification layer to the image recognition model. In addition, a method of classifying an input image and a system thereof are provided.

A method of generating an image recognition model for recognising an input image and a system thereof are provided. The method includes appending at least one feature extraction layer to the image recognition model, extracting a plurality of feature vectors from a set of predetermined images, grouping the plurality of feature vectors into a plurality of categories, clustering the plurality of feature vectors of each of the plurality of categories into at least one cluster, determining at least one centroid for each of the at least one cluster, such that each of the at least one cluster comprises at least one centroid, such that each of the at least one centroid is represented by a feature vector, generating a classification layer based on the feature vector of the at least one centroid of the plurality of categories, and appending the classification layer to the image recognition model. In addition, a method of classifying an input image and a system thereof are provided.

When it is assumed that a large-scale Deep Neural Network for autonomous driving applied compression, there are problems of a decrease in recognition accuracy of a post-compression Neural Network (NN) model and an increase in a compression design period, due to a large number of harmful or unnecessary training images (invalid training images). A training image selection unit B100 calculates an influence value on an inference, and generates an indexed training image set 1004-1 necessary for an NN compression design, by using the influence value. A neural network compression unit P200 notified of the result via a memory P300 compresses the NN.

A parameter optimization device 800 optimizes input CNN structure information and outputs optimized CNN structure information, and includes stride and dilation use layer detection means 811 for extracting stride and dilation parameter information for each convolution layer from the input CNN structure information, and stride and dilation use position modification means 812 for changing the stride and dilation parameter information of the convolution layer.

A learning processing device and method achieves learning of a lightweight model that is completed in a short amount of time. The learning processing device obtains a new, second learning model from an existing first learning model. An input unit acquires a first learning model generated in advance by learning a first learning data set, and an unpruned neural network (hereinafter, NN). An important parameter identification unit uses the first learning model and the NN to initialize a NN to be learned, and uses a second learning data set and the initialized NN to identify a degree of importance of parameters in a recognition process of the initialized NN. A new model generation unit carries out a pruning process for deleting parameters which are not important from the initialized NN, thereby generating a second NN; and a learning unit uses the second learning data set to learn the second NN.

Described herein are systems and methods for applying machine learning to telematics data to generate a unique vehicle fingerprint by periodically receiving telematics data generated at a plurality of sensors of a vehicle; standardizing the telematics data; aggregating the standardized telematics data; applying a trained machine learning model to embed the aggregated telematics data into a low-dimensional state; and generating a unique vehicle fingerprint, the vehicle fingerprint comprising a static component, a dynamic component, or both a static component and a dynamic component; including iterative repetition to update the dynamic component of the vehicle fingerprint.

Described herein are systems and methods for applying machine learning to telematics data to generate a unique vehicle fingerprint by periodically receiving telematics data generated at a plurality of sensors of a vehicle; standardizing the telematics data; aggregating the standardized telematics data; applying a trained machine learning model to embed the aggregated telematics data into a low-dimensional state; and generating a unique vehicle fingerprint, the vehicle fingerprint comprising a static component, a dynamic component, or both a static component and a dynamic component; including iterative repetition to update the dynamic component of the vehicle fingerprint.

In a method of optimizing a neural network model, first model information about a first neural network model is received. Device information about a first target device that is used to execute the first neural network model is received. An analysis whether the first neural network model is suitable for executing on the first target device is performed, based on the first model information, the device information, and at least one of a plurality of suitability determination algorithms. A result of the analysis is output such that the first model information and the result of the analysis are displayed on a screen.