Abstract: Systems and techniques are disclosed to reduce pilot overhead by providing common reference signals coded with cover codes that are orthogonal in time and frequency domains. Common reference signals that are coded by cover codes orthogonal in both domains can be de-spread in both the time and frequency domains for improved resolution and larger pull-in windows for both. Also disclosed is semi-uniform pilot spacing in both the frequency and time domains. In a time domain, a first pilot symbol pair is spaced by a first time interval from each other and a second pilot symbol pair is spaced by a second time interval from the first pair, the second interval being greater than the first. In a frequency domain, a first set of pilot symbols is densely placed in a selected frequency band and a second set of pilot symbols is sparsely placed surrounding and including the selected frequency band.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a neural network that has one or more batch normalized neural network layers for use by a quantized inference system. One of the methods includes receiving a first batch of training data; determining batch normalization statistics for the first batch of training data; determining a correction factor from the batch normalization statistics for the first batch of training data and the long-term moving averages of the batch normalization statistics; generating batch normalized weights from the floating point weights for the batch normalized first neural network layer, comprising applying the correction factor to the floating point weights of the batch normalized first neural network layer; quantizing the batch normalized weights; determining a gradient of an objective function; and updating the floating point weights using the gradient.

Abstract: An apparatus of operating a computational network is configured to determine a low-rank approximation for one or more layers of the computational network based at least in part on a set of residual targets. A set of candidate rank vectors corresponding to the set of residual targets may be determined. Each of the candidate rank vectors may be evaluated using an objective function. A candidate rank vector may be selected and used to determine the low rank approximation. The computational network may be compressed based on the low-rank approximation. In turn the computational network may be operated using the one or more compressed layers.

Abstract: A method of dynamically updating a feature database that contains features corresponding to a known target object includes providing an image, extracting a first set of features from within the captured image, and comparing the first set of features to the features stored in the feature database. If it is determined that the target object is present in the image then at least one of the extracted features of the first set that are not already included in the feature database are added to the feature database.

Abstract: An apparatus for learning a rank of an artificial neural network is configured to decompose a weight tensor into a first weight tensor and a second weight tensor. A set of rank selection parameters are applied to the first weight tensor and the second weight tensor to truncate the rank of the first weight tensor and the second weight tensor. The set of rank selection parameters are updated simultaneously with the weight tensors by averaging updates calculated for each rank selection parameter of the set of rank selection parameters.

Abstract: Systems and techniques are disclosed to reduce pilot overhead by providing common reference signals coded with cover codes that are orthogonal in time and frequency domains. Common reference signals that are coded by cover codes orthogonal in both domains can be de-spread in both the time and frequency domains for improved resolution and larger pull-in windows for both. Also disclosed is semi-uniform pilot spacing in both the frequency and time domains. In a time domain, a first pilot symbol pair is spaced by a first time interval from each other and a second pilot symbol pair is spaced by a second time interval from the first pair, the second interval being greater than the first. In a frequency domain, a first set of pilot symbols is densely placed in a selected frequency band and a second set of pilot symbols is sparsely placed surrounding and including the selected frequency band.

Abstract: Systems and techniques are disclosed to reduce pilot overhead by providing common reference signals coded with cover codes that are orthogonal in time and frequency domains. Common reference signals that are coded by cover codes orthogonal in both domains can be de-spread in both the time and frequency domains for improved resolution and larger pull-in windows for both. Also disclosed is semi-uniform pilot spacing in both the frequency and time domains. In a time domain, a first pilot symbol pair is spaced by a first time interval from each other and a second pilot symbol pair is spaced by a second time interval from the first pair, the second interval being greater than the first. In a frequency domain, a first set of pilot symbols is densely placed in a selected frequency band and a second set of pilot symbols is sparsely placed surrounding and including the selected frequency band.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training a neural network that has one or more batch normalized neural network layers for use by a quantized inference system. One of the methods includes receiving a first batch of training data; determining batch normalization statistics for the first batch of training data; determining a correction factor from the batch normalization statistics for the first batch of training data and the long-term moving averages of the batch normalization statistics; generating batch normalized weights from the floating point weights for the batch normalized first neural network layer, comprising applying the correction factor to the floating point weights of the batch normalized first neural network layer; quantizing the batch normalized weights; determining a gradient of an objective function; and updating the floating point weights using the gradient.

Abstract: Systems and techniques are disclosed to reduce pilot overhead by providing common reference signals coded with cover codes that are orthogonal in time and frequency domains. Common reference signals that are coded by cover codes orthogonal in both domains can be de-spread in both the time and frequency domains for improved resolution and larger pull-in windows for both. Also disclosed is semi-uniform pilot spacing in both the frequency and time domains. In a time domain, a first pilot symbol pair is spaced by a first time interval from each other and a second pilot symbol pair is spaced by a second time interval from the first pair, the second interval being greater than the first. In a frequency domain, a first set of pilot symbols is densely placed in a selected frequency band and a second set of pilot symbols is sparsely placed surrounding and including the selected frequency band.

Abstract: An apparatus for learning a rank of an artificial neural network is configured to decompose a weight tensor into a first weight tensor and a second weight tensor. A set of rank selection parameters are applied to the first weight tensor and the second weight tensor to truncate the rank of the first weight tensor and the second weight tensor. The set of rank selection parameters are updated simultaneously with the weight tensors by averaging updates calculated for each rank selection parameter of the set of rank selection parameters.

Abstract: Systems and techniques are disclosed to reduce pilot overhead by providing common reference signals coded with cover codes that are orthogonal in time and frequency domains. Common reference signals that are coded by cover codes orthogonal in both domains can be de-spread in both the time and frequency domains for improved resolution and larger pull-in windows for both. Also disclosed is semi-uniform pilot spacing in both the frequency and time domains. In a time domain, a first pilot symbol pair is spaced by a first time interval from each other and a second pilot symbol pair is spaced by a second time interval from the first pair, the second interval being greater than the first. In a frequency domain, a first set of pilot symbols is densely placed in a selected frequency band and a second set of pilot symbols is sparsely placed surrounding and including the selected frequency band.

Abstract: An apparatus of operating a computational network is configured to determine a low-rank approximation for one or more layers of the computational network based at least in part on a set of residual targets. A set of candidate rank vectors corresponding to the set of residual targets may be determined. Each of the candidate rank vectors may be evaluated using an objective function. A candidate rank vector may be selected and used to determine the low rank approximation. The computational network may be compressed based on the low-rank approximation. In turn the computational network may be operated using the one or more compressed layers.

Abstract: Methods, systems, computer-readable media, and apparatuses for image processing and utilization are presented. In some embodiments, an image containing at a face of a user may be obtained using a mobile device. An orientation of the face of the user within the image may be determined using the mobile device. The orientation of the face of the user may be determined using multiple stages: (a) a rotation stage for controlling a rotation applied to a portion of the image, to generate a portion of rotated image, and (b) an orientation stage for controlling an orientation applied to orientation-specific feature detection performed on the portion of rotated image. The determined orientation of the face of the user may be utilized as a control input to modify a display rotation of the mobile device.

Abstract: Techniques are described for wireless communication. One method includes identifying a plurality of channel responses corresponding to a plurality of channels. Each channel of the plurality of channels corresponds to a pairing of a transmit antenna with a receive antenna. Each channel response of the plurality of channel responses corresponds to a plurality of tone subsets. The method also includes selecting, for each channel of the plurality of channels, a subset of non-frequency domain components of the channel response for the channel, and transmitting, for at least one channel of the plurality of channels, at least one subset of channel state information (CSI). The at least one subset of CSI is based at least in part on at least one of the selected subsets of non-frequency domain components.

Abstract: A method of dynamically updating a feature database that contains features corresponding to a known target object includes providing an image, extracting a first set of features from within the captured image, and comparing the first set of features to the features stored in the feature database. If it is determined that the target object is present in the image then at least one of the extracted features of the first set that are not already included in the feature database are added to the feature database.

Abstract: Systems and techniques are disclosed to reduce pilot overhead by providing common reference signals coded with cover codes that are orthogonal in time and frequency domains. Common reference signals that are coded by cover codes orthogonal in both domains can be de-spread in both the time and frequency domains for improved resolution and larger pull-in windows for both. Also disclosed is semi-uniform pilot spacing in both the frequency and time domains. In a time domain, a first pilot symbol pair is spaced by a first time interval from each other and a second pilot symbol pair is spaced by a second time interval from the first pair, the second interval being greater than the first. In a frequency domain, a first set of pilot symbols is densely placed in a selected frequency band and a second set of pilot symbols is sparsely placed surrounding and including the selected frequency band.

Abstract: A method of dynamically updating a feature database that contains features corresponding to a known target object includes providing an image, extracting a first set of features from within the captured image, and comparing the first set of features to the features stored in the feature database. If it is determined that the target object is present in the image then at least one of the extracted features of the first set that are not already included in the feature database are added to the feature database.

Abstract: A method, a computer-readable medium, and an apparatus for compressing a neural network with an unlabeled data set are provided. The apparatus may generate a first set of consecutive layers for the neural network. The first set of consecutive layers may share inputs with a second set of consecutive layers of the neural network. The apparatus may adjust weights associated with the first set of consecutive layers based on a function the difference between a first set of output values from the first set of consecutive layers and a second set of output values from the second set of consecutive layers in response to the unlabeled data set. The apparatus may remove the second set of consecutive layers from the neural network when the function of the difference between the first set of output values and the second set of output values satisfies a threshold.

Abstract: Systems and techniques are disclosed to reduce pilot overhead by providing common reference signals coded with cover codes that are orthogonal in time and frequency domains. Common reference signals that are coded by cover codes orthogonal in both domains can be de-spread in both the time and frequency domains for improved resolution and larger pull-in windows for both. Also disclosed is semi-uniform pilot spacing in both the frequency and time domains. In a time domain, a first pilot symbol pair is spaced by a first time interval from each other and a second pilot symbol pair is spaced by a second time interval from the first pair, the second interval being greater than the first. In a frequency domain, a first set of pilot symbols is densely placed in a selected frequency band and a second set of pilot symbols is sparsely placed surrounding and including the selected frequency band.

Abstract: Methods, systems, and devices for wireless communication are described. One method includes identifying a plurality of intermediate precoders corresponding to a plurality of tone subsets. The plurality of intermediate precoders define a plurality of vectors across the plurality of tone subsets. The method further includes selecting, for each vector of the plurality of vectors, a subset of non-frequency domain components of the vector, such as time-domain components; modifying the plurality of intermediate precoders to a plurality of smoothed precoders based at least in part on the selected subset of non-frequency domain components for each vector; and precoding a plurality of transmit streams using the plurality of smoothed precoders. The plurality of smoothed precoders is smoothed in a frequency domain compared to the plurality of intermediate precoders. Smoothing precoders may enable application of wideband channel estimation techniques using user equipment (UE)-specific reference signals.