Abstract: An image encoder includes a processor and a memory. The memory includes instructions configured to cause the processor to perform operations. In one example implementation, the operations may include determining whether a dictionary item is available for replacing a block of an image being encoded, the determining based on a hierarchical lookup mechanism, and encoding the image along with reference information of the dictionary item in response to determining that the dictionary item is available. In one more example implementation, the operations may include performing principal component analysis (PCA) on a block to generate a corresponding projected block, the block being associated with a group of images, comparing the projected block with a corresponding threshold, descending the block recursively based on the threshold until a condition is satisfied, and identifying a left over block as a cluster upon satisfying of the condition.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for providing a machine learning model that is trained to perform a machine learning task. In one aspect, a method comprises receiving a request to train a machine learning model on a set of training examples; determining a set of one or more meta-data values characterizing the set of training examples; using a mapping function to map the set of meta-data values characterizing the set of training examples to data identifying a particular machine learning model architecture; selecting, using the particular machine learning model architecture, a final machine learning model architecture for performing the machine learning task; and training a machine learning model having the final machine learning model architecture on the set of training examples.

Abstract: Artificial image generation may include obtaining a source image, identifying quantization information from the source image, wherein identifying the quantization information includes identifying multiresolution quantization interval information from the source image, generating a restoration filtered image by restoration filtering the source image, generating a constrained restoration filtered image by constraining the restoration filtered image based on the quantization information, obtaining an unconstrained artificial image based on the constrained restoration filtered image and a generative artificial neural network obtained using a generative adversarial network, obtaining the artificial image by constraining the unconstrained artificial image based on the quantization information, and outputting the artificial image.

Abstract: A method for generating random numbers includes initializing a pseudo-random number generator (PRNG) having a state of 2048 bits comprising inner bits and outer bits, the inner bits comprising the first 128 bits of the 2048 bits and the outer bits comprising the remaining bits of the 2048 bits. The method also includes retrieving AES round keys from a key source, and for a threshold number of times, executing a round function using the AES round keys by XOR'ing odd-numbered branches of a Feistel network having 16 branches of 128 bits with a function of corresponding even-numbered neighbor branches of the Feistel network, and shuffling each branch of 128 bits into a prescribed order. The method also includes executing an XOR of the inner bits of the permuted state with the inner bits of a previous state.

Abstract: Systems, methods, and computer readable media related to generating query variants for a submitted query. In many implementations, the query variants are generated utilizing a generative model. A generative model is productive, in that it can be utilized to actively generate a variant of a query based on application of tokens of the query to the generative model, and optionally based on application of additional input features to the generative model.

Abstract: The present disclosure provides systems and methods for modification (e.g., pruning, compression, quantization, etc.) of artificial neural networks based on estimations of the utility of network connections (also known as “edges”). In particular, the present disclosure provides novel techniques for estimating the utility of one or more edges of a neural network in a fashion that requires far less expenditure of resources than calculation of the actual utility. Based on these estimated edge utilities, a computing system can make intelligent decisions regarding network pruning, network quantization, or other modifications to a neural network. In particular, these modifications can reduce resource requirements associated with the neural network. By making these decisions with knowledge of and based on the utility of various edges, this reduction in resource requirements can be achieved with only a minimal, if any, degradation of network performance (e.g., prediction accuracy).

Abstract: An apparatus includes a processor that is configured to select a palette entry in the palette for coding a value of a pixel of the image block; obtain respective palette indexes of neighboring pixels of the pixel; select, using the respective palette indexes, an entropy code for coding an index of the palette entry; and code the palette entry using the entropy code. A method includes obtaining respective palette indexes of neighboring pixels of a pixel of the image block; selecting an entropy code using the respective palette indexes; decoding, from a encoded bitstream, an index of a palette entry; and setting a value of the pixel using the palette entry.

Abstract: Systems, methods, and computer readable media related to generating query variants for a submitted query. In many implementations, the query variants are generated utilizing a generative model. A generative model is productive, in that it can be utilized to actively generate a variant of a query based on application of tokens of the query to the generative model, and optionally based on application of additional input features to the generative model.

Abstract: Methods are provided for improving the quality and compression factor of compressed images. The methods include determining frequency-band-specific quantization levels on a block-by-block level. This results in an adaptive dead zone, allowing certain blocks to be represented by fewer nonzero elements while other blocks are represented by more nonzero elements. Accordingly, the quality of the encoded image is improved while maintaining or improving the compression ratio. The adaptive quantization level is determined by comparing a post-quantization energy level to a threshold energy criterion for each frequency band within a block. Where the energy threshold criterion is not satisfied via these methods, additional methods can be applied to improve the image quality.

Abstract: A method for partitioning a block of an image to reduce quantization artifacts includes estimating an expected entropy of the block; partitioning the block into sub-blocks, where each sub-block having a size of a smallest possible partition size; calculating respective amounts of visual masking for the sub-blocks; selecting, as a visual masking characteristic of the block, a highest visual masking value of the respective amounts of visual masking for the sub-blocks; combining the visual masking characteristic of the block and the expected entropy of the block to obtain a splitting indicator value; and determining whether to split the block based on the splitting indicator.

Abstract: The loss of image quality during compression is controlled using a sequence of quality control metrics. The sequence of quality control metrics is selected for quantizing transform coefficients within an area of the image based on an error level definition. Candidate bit costs are then determined by quantizing the transform coefficients according to the error level definition or a modified error level and the sequence of quality control metrics. Where the candidate bit cost resulting from using the modified error level is lower than the candidate bit cost resulting from using the error level definition, the transform coefficients are quantized according to the modified error level and the sequence of quality control metrics. Otherwise, the transform coefficients are quantized based on the error level definition and according to the sequence of quality control metrics.

Abstract: A method for generating random numbers includes initializing a pseudo-random number generator (PRNG) having a state of 2048 bits comprising inner bits and outer bits, the inner bits comprising the first 128 bits of the 2048 bits and the outer bits comprising the remaining bits of the 2048 bits. The method also includes retrieving AES round keys from a key source, and for a threshold number of times, executing a round function using the AES round keys by XOR'ing odd-numbered branches of a Feistel network having 16 branches of 128 bits with a function of corresponding even-numbered neighbor branches of the Feistel network, and shuffling each branch of 128 bits into a prescribed order. The method also includes executing an XOR of the inner bits of the permuted state with the inner bits of a previous state.

Abstract: The present disclosure provides systems and methods for compressing and/or distributing machine learning models. In one example, a computer-implemented method is provided to compress machine-learned models, which includes obtaining, by one or more computing devices, a machine-learned model. The method includes selecting, by the one or more computing devices, a weight to be quantized and quantizing, by the one or more computing devices, the weight. The method includes propagating, by the one or more computing devices, at least a part of a quantization error to one or more non-quantized weights and quantizing, by the one or more computing devices, one or more of the non-quantized weights. The method includes providing, by the one or more computing devices, a quantized machine-learned model.

Abstract: The loss of image quality during compression is controlled using a sequence of quality control metrics. The sequence of quality control metrics is selected for quantizing transform coefficients within an area of the image based on an error level definition. Candidate bit costs are then determined by quantizing the transform coefficients according to the error level definition or a modified error level and the sequence of quality control metrics. Where the candidate bit cost resulting from using the modified error level is lower than the candidate bit cost resulting from using the error level definition, the transform coefficients are quantized according to the modified error level and the sequence of quality control metrics. Otherwise, the transform coefficients are quantized based on the error level definition and according to the sequence of quality control metrics.

Abstract: Video decoding may include transform coefficient continuity smoothing, which may include coefficient continuity smoothing, defined correlation coefficient smoothing, pixel range projection, and luminance correlated chrominance resampling. Coefficient continuity smoothing may include obtaining encoded block data from the encoded bitstream, the encoded block data corresponding to a current block from the reconstructed frame, and generating reconstructed block data for the current block based on the encoded block data using transform coefficient continuity smoothing.

Abstract: A method for generating random numbers includes initializing a pseudo-random number generator (PRNG) having a state of 2048 bits comprising inner bits and outer bits, the inner bits comprising the first 128 bits of the 2048 bits and the outer bits comprising the remaining bits of the 2048 bits. The method also includes retrieving AES round keys from a key source, and for a threshold number of times, executing a round function using the AES round keys by XOR'ing odd-numbered branches of a Feistel network having 16 branches of 128 bits with a function of corresponding even-numbered neighbor branches of the Feistel network, and shuffling each branch of 128 bits into a prescribed order. The method also includes executing an XOR of the inner bits of the permuted state with the inner bits of a previous state.

Abstract: Artificial image generation may include obtaining a source image, identifying quantization information from the source image, wherein identifying the quantization information includes identifying multiresolution quantization interval information from the source image, generating a restoration filtered image by restoration filtering the source image, generating a constrained restoration filtered image by constraining the restoration filtered image based on the quantization information, obtaining an unconstrained artificial image based on the constrained restoration filtered image and a generative artificial neural network obtained using a generative adversarial network, obtaining the artificial image by constraining the unconstrained artificial image based on the quantization information, and outputting the artificial image.

Abstract: A method includes obtaining respective filtered pixels for pixels of a reconstructed image; and obtaining an edge-preserved image using the respective filtered pixels. Obtaining the respective filtered pixels includes, for each pixel of the reconstructed image, obtaining a respective filtered pixel by selecting a pixel patch including the pixel and first neighboring pixels of the pixel that are at relative neighboring locations with respect to the pixel; calculating respective weights for the first neighboring pixels; and filtering the pixel using the respective weights of the first neighboring pixels and the neighboring pixels to obtain the respective filtered pixel.

Abstract: An image block is coded using entropy-inspired directional filtering. During encoding, intensity differences are determined for at least some pixels of an image block based on neighboring pixels of respective ones of the at least some pixels. Angles are estimated for each of those pixels based on the intensity differences. A main filtering direction of the image block is then determined based on the estimated angles. The image block is filtered according to the main filtering direction to remove artifacts along oblique edges associated with the image block. The filtered image block is then encoded to an encoded image. During decoding, an angular map indicating angles estimated for pixels of an encoded image block is received and used to determine the main filtering direction of the image block. The image block can then be filtered according to the main filtering direction and then output for display or storage.

Abstract: Artificial image generation may include obtaining a source image, identifying quantization information from the source image, wherein identifying the quantization information includes identifying multiresolution quantization interval information from the source image, generating a restoration filtered image by restoration filtering the source image, generating a constrained restoration filtered image by constraining the restoration filtered image based on the quantization information, obtaining an unconstrained artificial image based on the constrained restoration filtered image and a generative artificial neural network obtained using a generative adversarial network, obtaining the artificial image by constraining the unconstrained artificial image based on the quantization information, and outputting the artificial image.