Abstract: Systems and methods for data augmentation are described. Embodiments of the present disclosure receive a dataset that includes a plurality of nodes and a plurality of edges, wherein each of the plurality of edges connects two of the plurality of nodes; compute a first nonnegative matrix representing a homophilous cluster affinity; compute a second nonnegative matrix representing a heterophilous cluster affinity; compute a probability of an additional edge based on the dataset using a machine learning model that represents a homophilous cluster and a heterophilous cluster based on the first nonnegative matrix and the second nonnegative matrix; and generate an augmented dataset including the plurality of nodes, the plurality of edges, and the additional edge.

Abstract: One aspect of systems and methods for segment size estimation includes identifying a segment of users for a first time period based on time series data, wherein the time series data includes a series of interactions between users and a content channel and wherein the segment includes a portion of the users interacting with the content channel during the first time period; computing a segment return value for a second time period based on the time series data by computing a first subset and a second subset of the segment, wherein the first subset includes users that interact with the content channel greater than a threshold number of times during a range of the time series data and the second subset comprises a complement of the first subset with respect to the segment; and providing customized content to a user in the segment based on the segment return value.

Abstract: The application relates to an anti-back-transfer intake structure of a rotating detonation combustion chamber including a Tesla valve communicating with the rotating detonation combustion chamber and arranged at an inlet of the rotating detonation combustion chamber. The Tesla valve includes a casing and a flow passage, the casing is coaxially connected with an outer wall of the rotating detonation combustion chamber, the flow passage is arranged in the casing, and the flow passage has an inlet end for introducing air, and an outlet end connected with an annular passage of the rotating detonation combustion chamber.

Abstract: Methods and compositions for forming perovskite hole transport layers for use in manufacturing photovoltaic devices are described. Embodiments include using a plurality of hole transport materials to produce high-performance HTL contacts to improve performance and stability.

Abstract: Photovoltaic devices having contact layers are described herein. Devices, intermediate structures, and methods for making multilayer contacts for perovskite photovoltaic devices are provided. Embodiments include back contacts for N-I-P structures.

Abstract: The disclosure provides a displaying substrate, a manufacturing method thereof, and a display panel, and relates to the technical field of display. The displaying substrate comprises a first supporting base (1), plurality of vibrating element modules (2), and a display module (3). The display module (3) comprises display units (31), connecting units (32) and hollowed-out units (33). Each connecting unit (32) is located between two adjacent display units (31). Each hollowed-out unit (33) is located between two adjacent display units (31) except an area where the corresponding connecting unit (32) is located. The hollowed-out units (33) are provided with cavities (40) corresponding to the vibrating element modules (2). Orthographic projections of the hollowed-out units (33) on a reference plane cover orthographic projections of the vibrating element modules (2) on the reference plane. The vibrating element modules (2) and the cavities (40) form a transducer.

Abstract: The present disclosure relates to a backlight module, a method for designing the same, and a display device. The backlight module includes: a first substrate; a plurality of LED chips on the first substrate; and a light control structure on the first substrate. The backlight module includes a plurality of light control region groups in one-to-one correspondence with the plurality of light-emitting diode chips, each light control region group includes at least a first light control region and a second light control region. The light control structure includes a plurality of light control substructure groups respectively located in the plurality of light control region groups. Each light control substructure group includes at least a first light control substructure in the first light control region and a second light control substructure in the second light control region.

Abstract: This disclosure relates to a see-through display device, including: a collimated light source assembly, configured to form collimated light and control a light emitting direction; a first light extraction layer, configured to extract, in a collimated manner, the light ray transmitted inside the light guide plate through a light extraction outlet; an extinction layer and a second light extraction layer, wherein the extinction layer includes a light guide region and a light absorption region which are arranged alternately, and the second light extraction layer includes multiple light extraction inlets; a reflecting layer, arranged on a side, close to the light guide plate, of the second light extraction layer and configured to reflect the collimated light extracted from the light extraction outlet to the light guide plate; and a liquid crystal dimming layer. This disclosure further relates to a manufacturing method of the see-through display device.

Abstract: Developed and presented herein are embodiments of a new end-to-end approach for audio denoising, from a synthesis perspective. Instead of explicitly modelling the noise component in the input signal, embodiments directly synthesize the denoised audio from a generative model (or vocoder), as in text-to-speech systems. In one or more embodiments, to generate the phonetic contents for the autoregressive generative model, it is learned via a variational autoencoder with discrete latent representations. Furthermore, in one or more embodiments, a new matching loss is presented for the denoising purpose, which is masked on when the corresponding latent codes differ. As compared against other method on test datasets, embodiments achieve competitive performance and can be trained from scratch.

Abstract: Embodiments described herein provide methods and systems for facilitating actively-learned context modeling. In one embodiment, a subset of data is selected from a training dataset corresponding with an image to be compressed, the subset of data corresponding with a subset of data of pixels of the image. A context model is generated using the selected subset of data. The context model is generally in the form of a decision tree having a set of leaf nodes. Entropy values corresponding with each leaf node of the set of leaf nodes are determined. Each entropy value indicates an extent of diversity of context associated with the corresponding leaf node. Additional data from the training dataset is selected based on the entropy values corresponding with the leaf nodes. The updated subset of data is used to generate an updated context model for use in performing compression of the image.

Abstract: Embodiments provide systems, methods, and computer storage media for a Nonsymmetric Determinantal Point Process (NDPPs) for compatible set recommendations in a setting where data representing entities (e.g., items) arrives in a stream. A stream representing compatible sets of entities is received and used to update a latent representation of the entities and a compatibility distribution indicating likelihood of compatibility of subsets of the entities. The probability distribution is accessed in a single sequential pass to predict a compatible complete set of entities that completes an incomplete set of entities. The predicted complete compatible set is provided a recommendation for entities that complete the incomplete set of entities.

Abstract: The present application is applicable to the technical field of computer vision, and provides a method for reconstructing a three-dimensional object combining structured light and photometry and a terminal device, wherein the method comprises: acquiring N first images, wherein each first image is obtained by shooting after a coded pattern having a coding stripe sequence is projected to a three-dimensional object, and N is a positive integer; determining structured light depth information of the three-dimensional object based on the N first images; acquiring M second images, wherein the M second images are obtained by shooting after P light sources are respectively projected to the three-dimensional object from different directions, and M and P are positive integers; determining photometric information of the three-dimensional object based on the M second images; and reconstructing the three-dimensional object based on the structured light depth information and the photometric information.

Abstract: According to various embodiments, a method for encrypting image data for a neural network are disclosed. The method includes mixing the image data with other datapoints to form mixed data; and applying a pixel-wise random mask to the mixed data to form encrypted data. According to various embodiments, a method for encrypting text data for a neural network for natural language processing is disclosed. The method includes encoding each text datapoint via a pretrained text encoder to form encoded datapoints; mixing the encoded datapoints with other encoded datapoints to form mixed data; applying a random mask to the mixed data to form encrypted data; and incorporating the encrypted data into training a classifier of the neural network and fine-tuning the text encoder.

Abstract: Systems and methods for resource allocation are described. The systems and methods include receiving utilization data for computing resources shared by a plurality of users, updating a pricing agent using a reinforcement learning model based on the utilization data, identifying resource pricing information using the pricing agent, and allocating the computing resources to the plurality of users based on the resource pricing information.

Abstract: A control system facilitates active management of a streaming data system. Given historical data traffic for each data stream processed by a streaming data system, the control system uses a machine learning model to predict future data traffic for each data stream. The control system selects a matching between data streams and servers for a future time that minimizes a cost comprising a switching cost and a server imbalance cost based on the predicted data traffic for the future time. In some configurations, the matching is selected using a planning window comprising a number of future time steps dynamically selected based on uncertainty associated with the predicted data traffic. Given the selected matching, the control system may manage the streaming data system by causing data streams to be moved between servers based on the matching.

Abstract: WaveFlow is a small-footprint generative flow for raw audio, which may be directly trained with maximum likelihood. WaveFlow handles the long-range structure of waveform with a dilated two-dimensional (2D) convolutional architecture, while modeling the local variations using expressive autoregressive functions. WaveFlow may provide a unified view of likelihood-based models for raw audio, including WaveNet and WaveGlow, which may be considered special cases. It generates high-fidelity speech, while synthesizing several orders of magnitude faster than existing systems since it uses only a few sequential steps to generate relatively long waveforms. WaveFlow significantly reduces the likelihood gap that has existed between autoregressive models and flow-based models for efficient synthesis. Its small footprint with 5.91M parameters makes it 15 times smaller than some existing models. WaveFlow can generate 22.05 kHz high-fidelity audio 42.

Abstract: Developed and presented herein are embodiments of a new end-to-end approach for audio denoising, from a synthesis perspective. Instead of explicitly modelling the noise component in the input signal, embodiments directly synthesize the denoised audio from a generative model (or vocoder), as in text-to-speech systems. In one or more embodiments, to generate the phonetic contents for the autoregressive generative model, it is learned via a variational autoencoder with discrete latent representations. Furthermore, in one or more embodiments, a new matching loss is presented for the denoising purpose, which is masked on when the corresponding latent codes differ. As compared against other method on test datasets, embodiments achieve competitive performance and can be trained from scratch.

Abstract: Presented herein are embodiments of a non-autoregressive sequence-to-sequence model that converts text to an audio representation. Embodiment are fully convolutional, and a tested embodiment obtained about 46.7 times speed-up over a prior model at synthesis while maintaining comparable speech quality using a WaveNet vocoder. Interestingly, a tested embodiment also has fewer attention errors than the autoregressive model on challenging test sentences. In one or more embodiments, the first fully parallel neural text-to-speech system was built by applying the inverse autoregressive flow (IAF) as the parallel neural vocoder. System embodiments can synthesize speech from text through a single feed-forward pass. Also disclosed herein are embodiments of a novel approach to train the IAF from scratch as a generative model for raw waveform, which avoids the need for distillation from a separately trained WaveNet.

Abstract: WaveFlow is a small-footprint generative flow for raw audio, which may be directly trained with maximum likelihood. WaveFlow handles the long-range structure of waveform with a dilated two-dimensional (2D) convolutional architecture, while modeling the local variations using expressive autoregressive functions. WaveFlow may provide a unified view of likelihood-based models for raw audio, including WaveNet and WaveGlow, which may be considered special cases. It generates high-fidelity speech, while synthesizing several orders of magnitude faster than existing systems since it uses only a few sequential steps to generate relatively long waveforms. WaveFlow significantly reduces the likelihood gap that has existed between autoregressive models and flow-based models for efficient synthesis. Its small footprint with 5.91M parameters makes it 15 times smaller than some existing models. WaveFlow can generate 22.05 kHz high-fidelity audio 42.

Abstract: Presented herein are embodiments of a non-autoregressive sequence-to-sequence model that converts text to an audio representation. Embodiment are fully convolutional, and a tested embodiment obtained about 46.7 times speed-up over a prior model at synthesis while maintaining comparable speech quality using a WaveNet vocoder. Interestingly, a tested embodiment also has fewer attention errors than the autoregressive model on challenging test sentences. In one or more embodiments, the first fully parallel neural text-to-speech system was built by applying the inverse autoregressive flow (IAF) as the parallel neural vocoder. System embodiments can synthesize speech from text through a single feed-forward pass. Also disclosed herein are embodiments of a novel approach to train the IAF from scratch as a generative model for raw waveform, which avoids the need for distillation from a separately trained WaveNet.