A computing system includes at least one processor, and a memory communicatively coupled to the at least one processor. The processor is configured to receive at least two successive radar images of precipitation data, generate a motion vector field using the at least two successive radar images, forecast linear prediction imagery of future precipitation using the motion vector field, and generate corrected output imagery corresponding to the forecasted linear prediction imagery of the future precipitation corrected by a first neural network. In addition, the processor is further configured to receive, by a second neural network, the linear prediction imagery, and one of observed imagery and the corrected output imagery, and distinguish, by the second neural network, between the corrected output imagery and the observed imagery to produce conditioned output imagery. The processor is also configured to display the conditioned output imagery on a display.

A method of generating audio assets, comprising the steps of: receiving an input multi-layered audio asset comprising a plurality of audio layers, generating an input multi-channel image, wherein each channel of the input multi-channel image comprises an input image representative of one of the audio layers, training a generative model on the input multi-channel image and implementing the trained generative model to generate an output multi-channel image, wherein each channel of the output multi-channel image comprises an output image representative of an output audio layer, and generating an output multi-layered audio asset based on a combination of output audio layers derived from the output images.

Accuracy of a model extracting a graph structure as an intermediate representation from input data is improved. An encoding unit (100) extracts a feature amount of each of a plurality of vertices included in a graph structure (Tr) from input data (10), and calculates a likelihood that an edge is connected to the vertex. A sampling unit (130) determines the graph structure (Tr) based on a conversion result of a Gumbel-Softmax function for the likelihood. A learning unit (150) optimizes a decoding unit (140) and the encoding unit (100) by back propagation using a loss function including an error (L.sub.P) between output data (20) generated from the graph structure (Tr) and correct data.

Methods and systems for high-resolution image manipulation are disclosed. An original high-resolution image to be manipulated is obtained, as well as a driving signal indicating a manipulation result. The original high-resolution image is down-sampled to obtain a low-resolution image to be manipulated. Using a trained manipulation generator, a low-resolution manipulated image and a motion field are generated from the low-resolution image. The motion field represent pixel displacements of the low-resolution image to obtain the manipulation indicated by the driving signal. A high-frequency residual image is computed from the original high-resolution image. A high-frequency manipulated residual image is generated using the motion field. A high-resolution manipulated image is outputted by combining the high-frequency manipulated residual image and a low-frequency manipulated image generated from the low-resolution manipulated image by up-sampling.

According to some embodiments, systems and methods are provided, comprising receiving a code fragment exhibiting a static code issue; determining, via a trained exemption neural network, whether the received code fragment is exempt or not exempt from resolution; in a case it is not exempt, inputting the code fragment to a trained classification neural network; determining whether the static code issue is a syntactical static code issue or a non-syntactical static code issue; in a case it is a syntactical static code issue, inputting the code fragment to a first trained network to generate a first resolution; and in a case the static code issue is a non-syntactical static code issue, inputting the code fragment to a second trained network to generate a second resolution of the non-syntactical static code issue. Numerous other aspects are provided.

Described herein are means for integrating a defense mechanism into deep-learning-based systems to defend against adversarial attacks. For instance, an exemplary system is specially configured for adding a convolutinal defense layer to a neural network containing orthogonal kernels. Such a system generates the convolutional defense layer based on generating a set of learned kernals to increase diversity of network architecture, in which generating the set of learned kernals includes feeding an output of the convolutional defense layer into the neural network, further in which generating the convolutional defense layer includes selecting one or more orthogonal kernals, duplicating as needed and arranging them in a particular order. Such an embodiment further includes training the neural network with the added convolutional defense layer based on the increased diversity of network architecture; and defending against adverse attacks via constraining the effect of adversarial data generated by the adversarial attacks.

This disclosure describes methods, non-transitory computer readable storage media, and systems that generate realistic shading for three-dimensional objects inserted into digital images. The disclosed system utilizes a light encoder neural network to generate a representation embedding of lighting in a digital image. Additionally, the disclosed system determines points of the three-dimensional object visible within a camera view. The disclosed system generates a self-occlusion map for the digital three-dimensional object by determining whether fixed sets of rays uniformly sampled from the points intersects with the digital three-dimensional object. The disclosed system utilizes a generator neural network to determine a shading map for the digital three-dimensional object based on the representation embedding of lighting in the digital image and the self-occlusion map. Additionally, the disclosed system generates a modified digital image with the three-dimensional object inserted into the digital image with consistent lighting of the three-dimensional object and the digital image.

Method and system of training a student neural network using adversarial learning and knowledge distillation, including: training a generator to generate adversarial data samples for respective training data samples by masking parts of the training data samples with an objective of maximizing a divergence between output predictions generated by the student neural network and a teacher neural network model for the adversarial data samples; and training the student neural network based on objectives of (i) minimizing a divergence between output predictions generated by the student neural network and the teacher neural network model for the adversarial data samples, and (ii) minimizing a divergence between output predictions generated by the student neural network and the teacher neural network model for the training data samples.

Method and system of training a student neural network using adversarial learning and knowledge distillation, including: training a generator to generate adversarial data samples for respective training data samples by masking parts of the training data samples with an objective of maximizing a divergence between output predictions generated by the student neural network and a teacher neural network model for the adversarial data samples; and training the student neural network based on objectives of (i) minimizing a divergence between output predictions generated by the student neural network and the teacher neural network model for the adversarial data samples, and (ii) minimizing a divergence between output predictions generated by the student neural network and the teacher neural network model for the training data samples.

A system and method to capture and interact with a comprehensive digital record of an individual's biographical history and produce a synthetic model of their personality. The captured biographical history is a detailed record of this individual's actions, interactions, and experiences over a period which may span decades of their lifetime. The biographical history is indexed by areas of data variability and neural network confidence variability to identify points of likely human interest. A synthetic personality model is generated as a representation of the individual's personality structure, biases, sentiments, and traits. The synthetic personality can be interacted with through a digital interface and demonstrates the interaction patterns, triggers, and habits of the original individual. The functioning and the performance of the system over an individual's lifespan are optimized through data synthesis and disposition.