Abstract: Systems, apparatuses and methods may provide for conducting an importance measurement of a plurality of parameters in a trained neural network and setting a subset of the plurality of parameters to zero based on the importance measurement. Additionally, the pruned neural network may be re-trained. In one example, conducting the importance measurement includes comparing two or more parameter values that contain covariance matrix information.

Abstract: A semiconductor package apparatus may include technology to apply a trained scene text detection network to an image to identify a core text region, a supportive text region, and a background region of the image, and detect text in the image based on the identified core text region and supportive text region. Other embodiments are disclosed and claimed.

Abstract: A system to perform multi-modal analysis has at least three distinct characteristics: an early abstraction layer for each data modality integrating homogeneous feature cues coming from different deep learning architectures for that data modality, a late abstraction layer for further integrating heterogeneous features extracted from different models or data modalities and output from the early abstraction layer, and a propagation-down strategy for joint network training in an end-to-end manner. The system is thus able to consider correlations among homogeneous features and correlations among heterogenous features at different levels of abstraction. The system further extracts and fuses discriminative information contained in these models and modalities for high performance emotion recognition.

Abstract: Methods and systems are disclosed using improved Convolutional Neural Networks (CNN) for image processing. In one example, an input image is down-sampled into smaller images with a smaller resolution than the input image. The down-sampled smaller images are processed by a CNN having a last layer with a reduced number of nodes than a last layer of a full CNN used to process the input image at a full resolution. A result is outputted based on the processed down-sampled smaller images by the CNN having a last layer with a reduced number of nodes. In another example, shallow CNN networks are built randomly. The randomly built shallow CNN networks are combined to imitate a trained deep neural network (DNN).

Abstract: A mechanism is described for facilitating slimming of neural networks in machine learning environments. A method of embodiments, as described herein, includes learning a first neural network associated with machine learning processes to be performed by a processor of a computing device, where learning includes analyzing a plurality of channels associated with one or more layers of the first neural network. The method may further include computing a plurality of scaling factors to be associated with the plurality of channels such that each channel is assigned a scaling factor, wherein each scaling factor to indicate relevance of a corresponding channel within the first neural network. The method may further include pruning the first neural network into a second neural network by removing one or more channels of the plurality of channels having low relevance as indicated by one or more scaling factors of the plurality of scaling factors assigned to the one or more channels.

Abstract: Methods and systems are disclosed using camera devices for deep channel and Convolutional Neural Network (CNN) images and formats. In one example, image values are captured by a color sensor array in an image capturing device or camera. The image values provide color channel data. The captured image values by the color sensor array are input to a CNN having at least one CNN layer. The CNN provides CNN channel data for each layer. The color channel data and CNN channel data is to form a deep channel image that stored in a memory. In another example, image values are captured by sensor array. The captured image values by the sensor array are input a CNN having a first CNN layer. An output is generated at the first CNN layer using the captured image values by the color sensor array. The output of the first CNN layer is stored as a feature map of the captured image.

Abstract: Techniques are provided for neural network based, human attribute recognition, guided by anatomical key-points and statistic correlation models. Attributes include characteristics that can be visibly identified or inferred from an image, such as gender, hairstyle, clothing style, etc. A methodology implementing the techniques according to an embodiment includes applying an attribute feature extraction (AFE) convolutional neural network (CNN) to an image of a human to generate attribute feature maps based on the image. The method further includes applying a key-point guided proposal (KPG) CNN to the image of the human to generate proposed hierarchical regions of the image based on associated anatomical key-points.

Abstract: Systems, apparatuses and methods may provide for detecting a facial image including generating a spatial convolutional neural network score for one or more detected facial images from a facial image detector, generating a temporal convolutional network score for detected facial video frames from the facial image detector and generating a combined spatial-temporal score to determine whether a detected facial image gains user access to a protected resource.

Abstract: Techniques related to implementing convolutional neural networks for object recognition are discussed. Such techniques may include generating a set of binary neural features via convolutional neural network layers based on input image data and applying a strong classifier to the set of binary neural features to generate an object label for the input image data.

Abstract: Methods and apparatus are disclosed for enhancing a binary weight neural network using a dependency tree. A method of enhancing a convolutional neural network (CNN) having binary weights includes constructing a tree for obtained binary tensors, the tree having a plurality of nodes beginning with a root node in each layer of the CNN. A convolution is calculated of an input feature map with an input binary tensor at the root node of the tree. A next node is searched from the root node of the tree and a convolution is calculated at the next node using a previous convolution result calculated at the root node of the tree. The searching of a next node from root node is repeated for all nodes from the root node of the tree, and a convolution is calculated at each next node using a previous convolution result.

Abstract: One embodiment provides for a compute apparatus to perform machine learning operations, the compute apparatus comprising instruction decode logic to decode a single instruction including multiple operands into a single decoded instruction, the multiple operands including a first operand and a second operand, the first operand including vector of one-hot coded weights and the second operand including a vector of input data; and a general-purpose graphics compute unit including a first logic unit, the general-purpose graphics compute unit to execute the single decoded instruction, wherein to execute the single decoded instruction includes to perform multiple operations on the first set of operands and the second set of operands.

Abstract: A convolutional neural network framework is described that uses reverse connection and obviousness priors for object detection. A method includes performing a plurality of layers of convolutions and reverse connections on a received image to generate a plurality of feature maps, determining an objectness confidence for candidate bounding boxes based on outputs of an objectness prior, determining a joint loss function for each candidate bounding box by combining an objectness loss, a bounding box regression loss and a classification loss, calculating network gradients over positive boxes and negative boxes, updating network parameters within candidate bounding boxes using the joint loss function, repeating performing the convolutions through to updating network parameters until the training converges, and outputting network parameters for object detection based on the training images.

Abstract: Combinatorial shape regression is described as a technique for face alignment and facial landmark detection in images. As described stages of regression may be built for multiple ferns for a facial landmark detection system. In one example a regression is performed on a training set of images using face shapes, using facial component groups, and using individual face point pairs to learn shape increments for each respective image in the set of images. A fern is built based on this regression. Additional regressions are performed for building additional ferns. The ferns are then combined to build the facial landmark detection system.

Abstract: Methods and apparatus for discriminative semantic transfer and physics-inspired optimization in deep learning are disclosed. A computation training method for a convolutional neural network (CNN) includes receiving a sequence of training images in the CNN of a first stage to describe objects of a cluttered scene as a semantic segmentation mask. The semantic segmentation mask is received in a semantic segmentation network of a second stage to produce semantic features. Using weights from the first stage as feature extractors and weights from the second stage as classifiers, edges of the cluttered scene are identified using the semantic features.

Abstract: Methods and apparatus are disclosed for enhancing a neural network using binary tensor and scale factor pairs. For one example, a method of optimizing a trained convolutional neural network (CNN) includes initializing an approximation residue as a trained weight tensor for the trained CNN. A plurality of binary tensors and scale factor pairs are determined. The approximation residue is updated using the binary tensors and scale factor pairs.

Abstract: Methods and systems for budgeted and simplified training of deep neural networks (DNNs) are disclosed. In one example, a trainer is to train a DNN using a plurality of training sub-images derived from a down-sampled training image. A tester is to test the trained DNN using a plurality of testing sub-images derived from a down-sampled testing image. In another example, in a recurrent deep Q-network (RDQN) having a local attention mechanism located between a convolutional neural network (CNN) and a long-short time memory (LSTM), a plurality of feature maps are generated by the CNN from an input image. Hard-attention is applied by the local attention mechanism to the generated plurality of feature maps by selecting a subset of the generated feature maps. Soft attention is applied by the local attention mechanism to the selected subset of generated feature maps by providing weights to the selected subset of generated feature maps in obtaining weighted feature maps.

Abstract: Described herein are systems and methods for providing deeply stacked automated program synthesis. In one embodiment, an apparatus to perform automated program synthesis includes a memory to store instructions for automated program synthesis and a compute cluster coupled to the memory. The compute cluster supports the instructions for performing the automated program synthesis including partitioning sketched data into partitions, training diverse sets of individual program synthesis units each having different capabilities with partitioned sketched data and for each partition applying respective transformations, and generating sketched baseline data for each individual program synthesis unit.

Abstract: Methods and systems are disclosed for boosting deep neural networks for deep learning. In one example, in a deep neural network including a first shallow network and a second shallow network, a first training sample is processed by the first shallow network using equal weights. A loss for the first shallow network is determined based on the processed training sample using equal weights. Weights for the second shallow network are adjusted based on the determined loss for the first shallow network. A second training sample is processed by the second shallow network using the adjusted weights. In another example, in a deep neural network including a first weak network and a second weak network, a first subset of training samples is processed by the first weak network using initialized weights. A classification error for the first weak network on the first subset of training samples is determined.

Abstract: Methods and systems are disclosed using improved training and learning for deep neural networks. In one example, a deep neural network includes a plurality of layers, and each layer has a plurality of nodes. For each L layer in the plurality of layers, the nodes of each L layer are randomly connected to nodes in a L+1 layer. For each L+1 layer in the plurality of layers, the nodes of each L+1 layer are connected to nodes in a subsequent L layer in a one-to-one manner. Parameters related to the nodes of each L layer are fixed. Parameters related to the nodes of each L+1 layers are updated, and L is an integer starting with 1. In another example, a deep neural network includes an input layer, output layer, and a plurality of hidden layers. Inputs for the input layer and labels for the output layer are determined related to a first sample. Similarity between different pairs of inputs and labels between a second sample with the first sample is estimated using Gaussian regression process.

Abstract: Techniques are provided for facial image replacement between a reference facial image and a target facial image, of varying pose and illumination, using 3-dimensional morphable face models (3DMMs). A methodology implementing the techniques according to an embodiment includes fitting the reference face and the target face to a first and second 3DMM, respectively. The method further includes generating a texture map based on the fitted 3D reference face and rendering the fitted 3D reference face to a pose of the fitted 3D target face. The rendering is based on parameters of the first 3DMM, parameters of the second 3DMM, and the generated texture map associated with the fitted 3D reference face. The method further includes, determining a region of interest of the target facial image; and blending the rendered 3D reference face onto the region of interest of the target facial image to generate a replaced facial image.