Abstract: Methods, devices, systems, and instructions for adaptive quantization in an artificial neural network (ANN) calculate a distribution of ANN information; select a quantization function from a set of quantization functions based on the distribution; apply the quantization function to the ANN information to generate quantized ANN information; load the quantized ANN information into the ANN; and generate an output based on the quantized ANN information. Some examples recalculate the distribution of ANN information and reselect the quantization function from the set of quantization functions based on the resampled distribution if the output does not sufficiently correlate with a known correct output. In some examples, the ANN information includes a set of training data. In some examples, the ANN information includes a plurality of link weights.

Abstract: Methods, devices, systems, and instructions for adaptive quantization in an artificial neural network (ANN) calculate a distribution of ANN information; select a quantization function from a set of quantization functions based on the distribution; apply the quantization function to the ANN information to generate quantized ANN information; load the quantized ANN information into the ANN; and generate an output based on the quantized ANN information. Some examples recalculate the distribution of ANN information and reselect the quantization function from the set of quantization functions based on the resampled distribution if the output does not sufficiently correlate with a known correct output. In some examples, the ANN information includes a set of training data. In some examples, the ANN information includes a plurality of link weights.

Abstract: Systems, apparatuses, and methods for adjusting group sizes to match a processor lane width are described. In early iterations of an algorithm, a processor partitions a dataset into groups of data points which are integer multiples of the processing lane width of the processor. For example, when performing a K-means clustering algorithm, the processor determines that a first plurality of data points belong to a first group during a given iteration. If the first plurality of data points is not an integer multiple of the number of processing lanes, then the processor reassigns a first number of data points from the first plurality of data points to one or more other groups. The processor then performs the next iteration with these first number of data points assigned to other groups even though the first number of data points actually meets the algorithmic criteria for belonging to the first group.

Abstract: Methods, devices, systems, and instructions for adaptive quantization in an artificial neural network (ANN) calculate a distribution of ANN information; select a quantization function from a set of quantization functions based on the distribution; apply the quantization function to the ANN information to generate quantized ANN information; load the quantized ANN information into the ANN; and generate an output based on the quantized ANN information. Some examples recalculate the distribution of ANN information and reselect the quantization function from the set of quantization functions based on the resampled distribution if the output does not sufficiently correlate with a known correct output. In some examples, the ANN information includes a set of training data. In some examples, the ANN information includes a plurality of link weights.

Abstract: Systems, apparatuses, and methods for enhanced resolution video and security via machine learning are disclosed. A system is configured to receive a source code representation of a neural network. In one embodiment, the source code representation is a directed acyclic graph (DAG). The system determines if the source code representation includes any of one or more patterns, with each pattern including two or more adjacent layers. The system also identifies, for each pattern, a combined layer with which to replace the detected pattern. If any occurrences of the one or more patterns are detected in the source code representation, the system replaces each pattern with a corresponding combined layer. Additionally, the system generates an optimized representation of the neural network, wherein the optimized representation includes replacements for any detected patterns. The optimized representation can be utilized to generate an executable version of the neural network.

Abstract: A method and apparatus for performing a top-down Breadth-First Search (BFS) includes performing a first determination whether to convert to a bottom-up BFS. A second determination is performed whether to convert to the bottom-up BFS, based upon the first determination being positive. The bottom-up BFS is performed, based upon the first determination and the second determination being positive. A third determination is made whether to convert from the bottom-up BFS to the top-down BFS, based upon the third determination being positive.

Abstract: Systems, apparatuses, and methods for adjusting group sizes to match a processor lane width are described. In early iterations of an algorithm, a processor partitions a dataset into groups of data points which are integer multiples of the processing lane width of the processor. For example, when performing a K-means clustering algorithm, the processor determines that a first plurality of data points belong to a first group during a given iteration. If the first plurality of data points is not an integer multiple of the number of processing lanes, then the processor reassigns a first number of data points from the first plurality of data points to one or more other groups. The processor then performs the next iteration with these first number of data points assigned to other groups even though the first number of data points actually meets the algorithmic criteria for belonging to the first group.

Abstract: A method of multiplication of a sparse matrix and a vector to obtain a new vector and a system for implementing the method are claimed. Embodiments of the method are intended to optimize the performance of sparse matrix-vector multiplication in highly parallel processors, such as GPUs. The sparse matrix is stored in compressed sparse row (CSR) format.

Abstract: A method and apparatus for performing a top-down Breadth-First Search (BFS) includes performing a first determination whether to convert to a bottom-up BFS. A second determination is performed whether to convert to the bottom-up BFS, based upon the first determination being positive. The bottom-up BFS is performed, based upon the first determination and the second determination being positive. A third determination is made whether to convert from the bottom-up BFS to the top-down BFS, based upon the third determination being positive.

Abstract: A method of multiplication of a sparse matrix and a vector to obtain a new vector and a system for implementing the method are claimed. Embodiments of the method are intended to optimize the performance of sparse matrix-vector multiplication in highly parallel processors, such as GPUs. The sparse matrix is stored in compressed sparse row (CSR) format.