Facial expression recognition has been an active area of research over the past few decades, and it is still challenging due to high intra-class variation. Traditional approaches for this problem rely on hand-crafted features such as SIFT, HOG, LBP, followed by a classifier trained database images or videos. Most these works perform reasonably well datasets captured in controlled condition but fail more with image variation partial faces. In recent years, several proposed end-to-end...

Facial expression recognition has been an active research area over the past few decades, and it is still challenging due to high intra-class variation. Traditional approaches for this problem rely on hand-crafted features such as SIFT, HOG LBP, followed by a classifier trained database of images or videos. Most these works perform reasonably well datasets captured in controlled condition, but fail good more with image variation partial faces. In recent years, several proposed end-to-end...

With the popularity of social networks, and e-commerce websites, sentiment analysis has become a more active area research in past few years. On high level, tries to understand public opinion about specific product or topic, trends from reviews tweets. Sentiment plays an important role better understanding customer/user opinion, also extracting social/political trends. There been lot previous works for analysis, some based on hand-engineering relevant textual features, others different...

Iris recognition has drawn a lot of attention since the mid-twentieth century. Among all biometric features, iris is known to possess rich set features. Different features have been used perform in past. In this paper, two powerful sets are introduced be for recognition: scattering transform-based and textural PCA also applied on extracted reduce dimensionality feature vector while preserving most information its initial value. Minimum distance classifier template matching each new test...

Generating realistic biometric images has been an interesting and, at the same time, challenging problem. Classical statistical models fail to generate realistic-looking fingerprint images, as they are not powerful enough capture complicated texture representation in images. In this work, we present a machine learning framework based on generative adversarial networks (GAN), which is able sampled from prior distribution (learned set of training images). We also add suitable regularization...

Iris recognition has been an active research area during last few decades, because of its wide applications in security, from airports to homeland security border control. Different features and algorithms have proposed for iris the past. In this paper, we propose end-to-end deep learning framework based on residual convolutional neural network (CNN), which can jointly learn feature representation perform recognition. We train our model a well-known dataset using only training images each...

Fingerprint recognition has been utilized for cellphone authentication, airport security and beyond. Many different features algorithms have proposed to improve fingerprint recognition. In this paper, we propose an end-to-end deep learning framework using convolutional neural networks (CNNs) which can jointly learn the feature representation perform We train our model on a large-scale dataset, over previous approaches in terms of accuracy. Our is able achieve very high accuracy well-known...

Sparse decomposition has been extensively used for different applications including signal compression and denoising document analysis. In this paper, sparse is image segmentation. The proposed algorithm separates the background foreground using a sparse-smooth technique such that smooth components correspond to respectively. This tested on several test images from HEVC sequences shown have superior performance over other methods, as hierarchical k-means clustering in DjVu. segmentation can...

GPUs lack fundamental support for data-dependent parallelism and synchronization. While CUDA Dynamic Parallelism signals progress in this direction, many limitations challenges still remain. This paper introduces Wireframe, a hardware-software solution that enables generalized Wireframe applications to naturally express execution dependencies across different thread blocks through dependency graph abstraction at run-time, which is sent the GPU hardware kernel launch. At enforces specified...

The massive parallelism present in GPUs comes at the cost of reduced L1 and L2 cache sizes per thread, leading to serious contention problems such as thrashing. Hence, data access locality an application should be considered during thread scheduling improve execution time energy consumption. Recent works have tried use behavior regular structured applications scheduling, but difficult case irregular unstructured parallel remains explored. We PAVER , a P riority- A ware V ertex schedul ER...

Generating iris images which look realistic is both an interesting and challenging problem. Most of the classical statistical models are not powerful enough to capture complicated texture representation in images, therefore fail generate realistic. In this work, we present a machine learning framework based on generative adversarial network (GAN), able sampled from prior distribution (learned set training images). We apply two popular databases, very realistic, similar image those databases....

Runtime and scalability of large neural networks can be significantly affected by the placement operations in their dataflow graphs on suitable devices. With increasingly complex network architectures heterogeneous device characteristics, finding a reasonable is extremely challenging even for domain experts. Most existing automated approaches are impractical due to significant amount compute required inability generalize new, previously held-out graphs. To address both limitations, we...

Automatic prediction of age and gender from face images has drawn a lot attention recently, due it is wide applications in various facial analysis problems. However, to the large intra-class variation (such as lighting, pose, scale, occlusion), existing models are still behind desired accuracy level, which necessary for use these real-world applications. In this work, we propose deep learning framework, based on ensemble attentional residual convolutional networks, predict group with high...

Quantization has become a popular technique to compress neural networks and reduce compute cost, but most prior work focuses on studying quantization without changing the network size. Many real-world applications of have cost memory budgets, which can be traded off with model quality by number parameters. In this work, we use ResNet as case study systematically investigate effects inference cost-quality tradeoff curves. Our results suggest that for each bfloat16 model, there are quantized...

Face recognition has been an active research area in the past few decades. In general, face can be very challenging due to variations viewpoint, illumination, facial expression, etc. Therefore it is essential extract features which are invariant some or all of these variations. Here a new image representation, called scattering trans-form/network, used from faces. The transform kind convolutional network provides powerful multi-layer representation for signals. After extraction features, PCA...

Most compilers for machine learning (ML) frameworks need to solve many correlated optimization problems generate efficient code. Current ML rely on heuristics based algorithms these one at a time. However, this approach is not only hard maintain but often leads sub-optimal solutions especially newer model architectures. Existing approaches in the literature are sample inefficient, tackle single problem, and do generalize unseen graphs making them infeasible be deployed practice. To address...

The Register File (RF) is a critical structure in Graphics Processing Units (GPUs) responsible for large portion of the area and power. To simplify architecture RF, it organized multi-bank configuration with single port each bank. Not surprisingly, frequent accesses to register file during kernel execution incur sizeable overhead GPU power consumption, introduce delays as are serialized when conflicts occur. In this paper, we observe that there high degree temporal locality registers: within...

Palmprint is one of the most useful physiological biometrics that can be used as a powerful means in personal recognition systems. The major features palmprints are palm lines, wrinkles and ridges, many approaches use them different ways towards solving palmprint problem. Here we have proposed to set statistical wavelet-based features; capture general characteristics palmprints; find those information not evident spatial domain. Also two classification approaches, minimum distance classifier...