Abstract: A method is provided. The method includes selecting a neural network model, wherein the neural network model includes a plurality of layers, and wherein each of the plurality of layers includes weights and activations; modifying the neural network model by inserting a plurality of quantization layers within the neural network model; associating a cost function with the modified neural network model, wherein the cost function includes a first coefficient corresponding to a first regularization term, and wherein an initial value of the first coefficient is pre-defined; and training the modified neural network model to generate quantized weights for a layer by increasing the first coefficient until all weights are quantized and the first coefficient satisfies a pre-defined threshold, further including optimizing a weight scaling factor for the quantized weights and an activation scaling factor for quantized activations, and wherein the quantized weights are quantized using the optimized weight scaling factor.

Abstract: An apparatus and a method. The apparatus includes a receiver including an input for receiving a codeword of length mj, where m and j are each an integer; a processor configured to determine a decoding node tree structure with mj leaf nodes for the received codeword and receive an integer i indicating a level at which parallelism of order m is applied to the decoding node tree structure; and m successive cancellation decoders (SCDs) configured to decode, in parallel, each child node in the decoding node tree structure at level i.

Abstract: An image denoising neural network training architecture includes an image denoising neural network and a clean data neural network, and the image denoising neural network and clean data neural network share information between each other.

Abstract: Apparatuses and methods of manufacturing same, systems, and methods are described. In one aspect, a method includes generating a convolutional neural network (CNN) by training a CNN having three or more convolutional layers, and performing cascade training on the trained CNN. The cascade training includes an iterative process of one or more stages, in which each stage includes inserting a residual block (ResBlock) including at least two additional convolutional layers and training the CNN with the inserted ResBlock.

Abstract: A method and apparatus are provided. The method includes configuring a plurality of long short term memory (LSTM) networks, wherein each of the plurality of LSTM networks is at a different network layer, configuring a plurality of memory cells in a spatial domain of the plurality of LSTM networks, configuring the plurality of memory cells in a temporal domain of the plurality of LSTM networks, controlling an output of each of the plurality of LSTM networks based on highway connections to outputs from at least one previous layer and at least one previous time of the plurality of LSTM networks, and controlling the plurality of memory cells based on highway connections to memory cells from the at least one previous time.

Abstract: Detecting objects in an image includes: extracting core instance features from the image; calculating feature maps at multiscale resolutions from the core instance features; calculating detection boxes from the core instance features; calculating segmentation masks for each detection box of the detection boxes at the multiscale resolutions of the feature maps; merging the segmentation masks at the multiscale resolutions to generate an instance mask for each object detected in the image; refining the confidence scores of the merged segmentation masks by auxiliary networks calculating pixel level metrics; and outputting the instance masks as the detected objects.

Abstract: A method and system for decoding a signal are provided. The method includes receiving a signal, where the signal includes at least one symbol; decoding the signal in stages, where each at least one symbol is decoded into at least one bit per stage, wherein a Log-Likelihood Ratio (LLR) and a path metric are determined for each possible path for each at least one bit at each stage; determining the magnitudes of the LLRs; identifying K bits of the signal with the smallest corresponding LLR magnitudes; identifying, for each of the K bits, L possible paths with the largest path metrics at each decoder stage for a user-definable number of decoder stages; performing forward and backward traces, for each of the L possible paths, to determine candidate codewords; performing a Cyclic Redundancy Check (CRC) on the candidate codewords, and stopping after a first candidate codeword passes the CRC.

Abstract: An apparatus and a method. The apparatus includes a plurality of polarization processors, including n inputs and n outputs, where n is an integer, wherein the plurality of polarization processors is configured to polarize channels with different bit-channel reliability; and at least one permutation processor, including n inputs and n outputs, wherein each of the at least one permutation processor is connected between two of the plurality of polarization processors, and connects the n outputs of a first of the two of the plurality of polarizations processors to the n inputs of a second of the two of the plurality of polarization processors between which each of the at least one permutation processor is connected in a permutation pattern, wherein at least one permutation processor is configured to not further polarize a bit channel.

Abstract: A method, apparatus, and non-transitory computer-readable recording medium for generating an algebraic Spatially-Coupled Low-Density Parity-Check (SC LDPC) code are provided. The method includes selecting an LDPC block code over a finite field GF(q) with a girth of at least 6; constructing a parity-check matrix H from the selected LDPC block code; replicating H a user-definable number of times to form a two-dimensional array Hrep; constructing a masking matrix W with a user-definable spatially-coupled pattern; and masking a sub-matrix of Hrep using W to obtain a spatially-coupled parity-check matrix HSC, wherein a null space of HSC is the algebraic SC LDPC code.

Abstract: An apparatus and a method. The apparatus includes a plurality of polarization processors, including n inputs and n outputs, where n is an integer, wherein the plurality of polarization processors is configured to polarize channels with different bit-channel reliability; and at least one permutation processor, including n inputs and n outputs, wherein each of the at least one permutation processor is connected between two of the plurality of polarization processors, and connects the n outputs of a first of the two of the plurality of polarizations processors to the n inputs of a second of the two of the plurality of polarization processors between which each of the at least one permutation processor is connected in a permutation pattern, wherein at least one permutation processor is configured to not further polarize a bit channel.

Abstract: An apparatus and a method. The apparatus includes a receiver to receive a polar codeword of length mj; a processor configured to determine a decoding node tree structure with mj leaf nodes for the received codeword, and receive i indicating a level at which parallelism of order m is applied to the decoding node tree structure, wherein i indicates levels of the decoding node tree structure, and wherein the mj leaf nodes are at level j; and m successive cancellation list decoders (SCLDs) applied to each child node of each node in the decoding node tree structure at level i?1, wherein each of the m SCLDs executes in parallel to determine log likelihood ratios (LLRs) for a codeword of length mj-i, and wherein each of the m SCLDs uses LLRs of an associated parent node without using a hard decision or a soft reliability estimate of any other node of the other m SCLDs.

Abstract: An apparatus, a method, a method of manufacturing and apparatus, and a method of constructing an integrated circuit are provided. The apparatus includes a teacher network; a student network; a plurality of knowledge bridges between the teacher network and the student network, where each of the plurality of knowledge bridges provides a hint about a function being learned, and where a hint includes a mean square error or a probability; and a loss function device connected to the plurality of knowledge bridges and the student network. The method includes training a teacher network; providing hints to a student network by a plurality of knowledge bridges between the teacher network and the student network; and determining a loss function from outputs of the plurality of knowledge bridges and the student network.

Abstract: A system to recognize objects in an image includes an object detection network outputs a first hierarchical-calculated feature for a detected object. A face alignment regression network determines a regression loss for alignment parameters based on the first hierarchical-calculated feature. A detection box regression network determines a regression loss for detected boxes based on the first hierarchical-calculated feature. The object detection network further includes a weighted loss generator to generate a weighted loss for the first hierarchical-calculated feature, the regression loss for the alignment parameters and the regression loss of the detected boxes. A backpropagator backpropagates the generated weighted loss.

Abstract: A computing system includes: an inter-device interface configured to access a destination signal including an information portion for representing a content and an error-handling portion for describing the information portion relative to the content; a communication unit, coupled to the inter-device interface, configured to: generate a parity-check parameter based on a sparse configuration from the destination signal, and estimate the content based on decoding the information portion using the error-handling portion and the parity-check parameter.

Abstract: Disclosed herein is convolutional neural network (CNN) system for generating a classification for an input image. According to an embodiment, the CNN system comprises a sequence of neural network layers configured to: derive a feature map based on at least the input image; puncture at least one selection among the feature map and a kernel by setting the value of one or more elements of a row of the at least one selection to zero according to a pattern and cyclic shifting the pattern by a predetermined interval per row to set the value of one or more elements of the rest of the rows of the at least one selection according to the cyclic shifted pattern; convolve the feature map with the kernel to generate a first convolved output; and generate the classification for the input image based on at least the first convolved output.

Abstract: In a method for super resolution imaging, the method includes: receiving, by a processor, a low resolution image; generating, by the processor, an intermediate high resolution image having an improved resolution compared to the low resolution image; generating, by the processor, a final high resolution image based on the intermediate high resolution image and the low resolution image; and transmitting, by the processor, the final high resolution image to a display device for display thereby.

Abstract: Apparatuses and methods of manufacturing same, systems, and methods for generating a convolutional neural network (CNN) are described. In one aspect, a minimal CNN having, e.g., three or more layers is trained. Cascade training may be performed on the trained CNN to insert one or more intermediate layers until a training error is less than a threshold. When cascade training is complete, cascade network trimming of the CNN output from the cascade training may be performed to improve computational efficiency. To further reduce network parameters, convolutional filters may be replaced with dilated convolutional filters with the same receptive field, followed by additional training/fine-tuning.

Abstract: A electronic system includes: a support chip configured to receive an input code stream; a circular Viterbi mechanism, coupled to the support chip, configured to: generate a final path metric for the input code stream, store intermediate path metrics at the repetition depth, generate a repetition path metric for the input code stream, and calculate a soft correlation metric based on the final path metric, the repetition path metric, and the intermediate path metrics.

Abstract: An apparatus and a method. The apparatus includes a plurality of polarization processors, including n inputs and n outputs, where n is an integer; and at least one permutation processor, including n inputs and n outputs, wherein each of the at least one permutation processor is connected between two of the plurality of polarization processors, and connects the n outputs of a first of the two of the plurality of polarizations processors to the n inputs of a second of the two of the plurality of polarization processors between which each of the at least one permutation processor is connected in a permutation pattern that maximally polarizes the n outputs of the second of the two of the plurality of polarization processors.

Abstract: A system to recognize objects in an image includes an object detection network outputs a first hierarchical-calculated feature for a detected object. A face alignment regression network determines a regression loss for alignment parameters based on the first hierarchical-calculated feature. A detection box regression network determines a regression loss for detected boxes based on the first hierarchical-calculated feature. The object detection network further includes a weighted loss generator to generate a weighted loss for the first hierarchical-calculated feature, the regression loss for the alignment parameters and the regression loss of the detected boxes. A backpropagator backpropagates the generated weighted loss.