Art Convolutional Neural Networks Research Articles

Cross-Layer Design Exploration for Energy-Quality Tradeoffs in Spiking and Non-Spiking Deep Artificial Neural Networks

Deep learning convolutional artificial neural networks have achieved success in a large number of visual processing tasks and are currently utilized for many real-world applications like image search and speech recognition among others. However, despite achieving high accuracy in such classification problems, they involve significant computational resources. Over the past few years, non-spiking deep convolutional artificial neural network models have evolved into more biologically realistic and event-driven spiking deep convolutional artificial neural networks. Recent research efforts have been directed at developing mechanisms to convert traditional non-spiking deep convolutional artificial neural networks to the spiking ones where neurons communicate by means of spikes. However, there have been limited studies providing insights on the specific power, area, and energy benefits offered by the spiking deep convolutional artificial neural networks in comparison to their non-spiking counterparts. We perform a comprehensive study for hardware implementation of spiking/non-spiking deep convolutional artificial neural networks on MNIST, CIFAR10, and SVHN datasets. To this effect, we design AccelNN - a Neural Network Accelerator to execute neural network benchmarks and analyze the effects of circuit-architecture level techniques to harness event-drivenness. A comparative analysis between spiking and non-spiking versions of deep convolutional artificial neural networks is presented by performing trade-offs between recognition accuracy and corresponding power, latency and energy requirements.

Predicting weather forecast uncertainty with machine learning

Weather forecasts are inherently uncertain. Therefore, for many applications forecasts are only considered valuable if an uncertainty estimate can be assigned to them. Currently, the best method to provide a confidence estimate for individual forecasts is to produce an ensemble of numerical weather simulations, which is computationally very expensive. Here, we assess whether machine learning techniques can provide an alternative approach to predict the uncertainty of a weather forecast given the large‐scale atmospheric state at initialization. We propose a method based on deep learning with artificial convolutional neural networks that is trained on past weather forecasts. Given a new weather situation, it assigns a scalar value of confidence to medium‐range forecasts initialized from the said atmospheric state, indicating whether the predictability is higher or lower than usual for the time of the year. While our method has a lower skill than ensemble weather forecast models in predicting forecast uncertainty, it is computationally very efficient and outperforms a range of alternative methods that do not involve performing numerical forecasts. This shows that it is possible to use machine learning in order to estimate future forecast uncertainty from past forecasts. The main constraint in the performance of our method seems to be the number of past forecasts available for training the machine learning algorithm.

A convolutional neural network with feature fusion for real-time hand posture recognition

Gesture based human–computer interaction is both intuitive and versatile, with diverse applications such as in smart houses, operating theaters and vehicle infotainment systems. This paper presents a novel architecture, combining a convolutional neural network (CNN) and traditional feature extractors, capable of accurate and real-time hand posture recognition. The proposed architecture is evaluated on three distinct benchmark datasets and compared with the state-of-the art convolutional neural networks. Extensive experimentation is conducted using binary, grayscale and depth data, as well as two different validation techniques. The proposed feature fusion-based convolutional neural network (FFCNN) is shown to perform better across combinations of validation techniques and image representation. The recognition rate of FFCNN on binary images is equivalent to grayscale and depth when the aspect ratio of gestures is preserved. A real-time recognition system is presented with a demonstration video.

Automated skin melanoma diagnostics based on mathematical model of artificial convolutional neural network

In the last 10 years there has been a revolu on in the fi eld of computer image analysis and pa ern recogni on. Modern algorithms of computer vision equaled and even in some problems surpassed human capabili es. This jerk is largely due to the emergence and development of the technology of deep convolu onal neural networks.Recent developments in the fi eld of image processing and machine learning open up the prospect of crea ng systems based on ar fi cial neural convolu onal networks, superior to humans in problems of image classifi ca on, in par cular, in solving problems of analysis of various medical images. Among the most promising applica ons: automated recogni on and classifi ca on of skin diseases, detec on of pathologies on X-ray, CT, MRI, ultrasound imaging. In the proposed project, we will focusour a en on on the diagnosis of human skin diseases.At the moment, melanoma is one of the most dangerous types of malignant tumors of the skin with a lot of deaths due to rapid metastasis, which is difficult to treat. The development of computer vision technology has allowed the development of technical vision systems that allow detec on and classifi ca on of skin diseases with a quality that is comparable and in some cases exceeds the values a ained by man.In this paper, the authors propose an algorithm for the primary diagnosis of skin melanoma based on deep neural networks, achieving an accuracy of 91% for melanoma in dermatoscopic images. At the moment, the algorithm is implemented programma cally and is used in the test version of the online system for detec ng and classifying skin diseases, available at skincheckup.online.Thanks to this development, the prospect of a signifi cantincrease in the propor on of people subjected to preven ve examina on for the presence of skin diseases opens up. Along with this, an addi onal source of informa on for specialized professionals can also play a role in seng the right diagnosis.

Effective and High Computing Algorithms for Convolution Neural Networks

Training a large set of data takes GPU days using Deep convolution neural networks which are a time taking process. Self-driving cars require very low latency for pedestrian detection. Image recognition constrained by limited processing resources for mobile phones. The computation speed of the training set determines that in these situations convolution neural networks was a success. For large filters, Conventional Faster Fourier Transform based convolution is preferably fast, yet in case of small, 3 × 3 filters state of the art convolutional neural networks is used. By using Winograd's minimal filtering algorithms the new class of fast algorithms for convolutional neural networks was introduced by us. Instead of small tiles, minimal complexity convolution was computed by the algorithms, this increases the computing speed with small batch sizes and small filters. With the VGG network, we benchmark a GPU implementation of our algorithm and at batch sizes from 1 to 64 state of the art throughput was shown.

OPTIMAL UNIFORM QUANTIZATION OF PARAMETERS OF CONVOLUTIONAL NEURAL NETWORKS

State of the art convolutional neural networks provide high accuracy in solving a wide range of problems. Usually it is achieved by a significant increasing their computational complexity and the representation of the network parameters in single-precision floating point numbers. However, due to the limited resources, the application of networks in embedded systems and mobile applications in real time is problematic. One of the methods to solve this problem is to reduce the bit depth of data and use integer arithmetic. For this purpose, the network parameters are quantized. Performing quantization, it is necessary to ensure a minimum loss of recognition accuracy. The article proposes to use an optimal uniform quantizer with an adaptive step. The quantizer step depends on the distribution function of the quantized parameters. It reduces the effect of the quantization error on the recognition accuracy. There are also described approaches to improving the quality of quantization. The proposed quantization method is estimated on the CIFAR-10 database. It is shown that the optimal uniform quantizer for CIFAR-10 database with 8-bit representation of network parameters allows to achieve the accuracy of the initial trained network.

Neural network identification of people hidden from view with a single-pixel, single-photon detector

Light scattered from multiple surfaces can be used to retrieve information of hidden environments. However, full three-dimensional retrieval of an object hidden from view by a wall has only been achieved with scanning systems and requires intensive computational processing of the retrieved data. Here we use a non-scanning, single-photon single-pixel detector in combination with a deep convolutional artificial neural network: this allows us to locate the position and to also simultaneously provide the actual identity of a hidden person, chosen from a database of people (N = 3). Artificial neural networks applied to specific computational imaging problems can therefore enable novel imaging capabilities with hugely simplified hardware and processing times.

Large-Scale, High-Resolution Comparison of the Core Visual Object Recognition Behavior of Humans, Monkeys, and State-of-the-Art Deep Artificial Neural Networks

Primates, including humans, can typically recognize objects in visual images at a glance despite naturally occurring identity-preserving image transformations (e.g., changes in viewpoint). A primary neuroscience goal is to uncover neuron-level mechanistic models that quantitatively explain this behavior by predicting primate performance for each and every image. Here, we applied this stringent behavioral prediction test to the leading mechanistic models of primate vision (specifically, deep, convolutional, artificial neural networks; ANNs) by directly comparing their behavioral signatures against those of humans and rhesus macaque monkeys. Using high-throughput data collection systems for human and monkey psychophysics, we collected more than one million behavioral trials from 1472 anonymous humans and five male macaque monkeys for 2400 images over 276 binary object discrimination tasks. Consistent with previous work, we observed that state-of-the-art deep, feedforward convolutional ANNs trained for visual categorization (termed DCNNIC models) accurately predicted primate patterns of object-level confusion. However, when we examined behavioral performance for individual images within each object discrimination task, we found that all tested DCNNIC models were significantly nonpredictive of primate performance and that this prediction failure was not accounted for by simple image attributes nor rescued by simple model modifications. These results show that current DCNNIC models cannot account for the image-level behavioral patterns of primates and that new ANN models are needed to more precisely capture the neural mechanisms underlying primate object vision. To this end, large-scale, high-resolution primate behavioral benchmarks such as those obtained here could serve as direct guides for discovering such models.SIGNIFICANCE STATEMENT Recently, specific feedforward deep convolutional artificial neural networks (ANNs) models have dramatically advanced our quantitative understanding of the neural mechanisms underlying primate core object recognition. In this work, we tested the limits of those ANNs by systematically comparing the behavioral responses of these models with the behavioral responses of humans and monkeys at the resolution of individual images. Using these high-resolution metrics, we found that all tested ANN models significantly diverged from primate behavior. Going forward, these high-resolution, large-scale primate behavioral benchmarks could serve as direct guides for discovering better ANN models of the primate visual system.

403 A HIGHLY SENSITIVE AND HIGHLY SPECIFIC CONVOLUTIONAL NEURAL NETWORK-BASED ALGORITHM FOR AUTOMATED DIAGNOSIS OF ANGIODYSPLASIA IN SMALL BOWEL CAPSULE ENDOSCOPY

Capsule endoscopy (CE) has become a standard non-invasive tool for small bowel (SB) examination. However, with an average number of 50,000 SB still frames per CE video, lesions can be missed and CE reading remains a time-consuming activity. Therefore, the development of computer-aided algorithms for lesions’ detection has become an active research area in CE. Gastro-intestinal angiodysplasias (AGD) are the most common SB vascular lesions with an inherent risk of bleeding. This study aimed to develop a computer-assisted diagnosis (CAD) tool for SB-AGD detection in CE. A French national database (CAD-CAP) was created: 5296 deidentified third generation SB-CE videos (Pillcam® SB3 system, Medtronic) were collected from 13 centers; 5427 pathological still frames were extracted from 1369 SB-CE with at least one pathological findings; 20,000 normal (control) still frames were extracted from 200 normal SB-CE videos. Six hundred still images with typical AGD were selected from the database by expert readers. Six hundred normal still frames were randomly extracted as controls. The 600 selected AGD were annotated on each individual pathological frame, as “ground truth” reference. Two sets of 600 SB-CE still frames (300 normal frames and 300 frames with AGD, per set) were created. Different learning-based algorithms were tested including color-based approaches and deep feature extraction. The first set of still frames was used for the machine learning process and the second bank for the validation process. Among the tested algorithms, the one with the best diagnostic performance characteristics used a semantic segmentation images approach associated with an artificial convolutional neural network (CNN) for deep feature extraction (figure 1 from left to right: original image, annotated image, CNN approach, CNN approach on original image). This algorithm reached a 100% sensitivity, a 95,83% specificity, a 96.15% positive predictive value, and a 100% negative predictive value for AGD detection. Reproducibility was perfect (Kappa = 1.0). This study shows that a computed algorithm based on a CNN has very good reproducibility, diagnostic sensitivity (100%), and specificity (95.83%), for detection of AGD on still frames. This study opens the way to automated SB-CE reading softwares, and calls for prospective evaluations of CNN at the video level.

OCT-based deep learning algorithm for the evaluation of treatment indication with anti-vascular endothelial growth factor medications.

Intravitreal injections with anti-vascular endothelial growth factor (anti-VEGF) medications have become the standard of care for their respective indications. Optical coherence tomography (OCT) scans of the central retina provide detailed anatomical data and are widely used by clinicians in the decision-making process of anti-VEGF indication. In recent years, significant progress has been made in artificial intelligence and computer vision research. We trained a deep convolutional artificial neural network to predict treatment indication based on central retinal OCT scans without human intervention. A total of 183,402 retinal OCT B-scans acquired between 2008 and 2016 were exported from the institutional image archive of a university hospital. OCT images were cross-referenced with the electronic institutional intravitreal injection records. OCT images with a following intravitreal injection during the first 21days after image acquisition were assigned into the 'injection' group, while the same amount of random OCT images without intravitreal injections was labeled as 'no injection'. After image preprocessing, OCT images were split in a 9:1 ratio to training and test datasets. We trained a GoogLeNet inception deep convolutional neural network and assessed its performance on the validation dataset. We calculated prediction accuracy, sensitivity, specificity, and receiver operating characteristics. The deep convolutional neural network was successfully trained on the extracted clinical data. The trained neural network classifier reached a prediction accuracy of 95.5% on the images in the validation dataset. For single retinal B-scans in the validation dataset, a sensitivity of 90.1% and a specificity of 96.2% were achieved. The area under the receiver operating characteristic curve was 0.968 on a per B-scan image basis, and 0.988 by averaging over six B-scans per examination on the validation dataset. Deep artificial neural networks show impressive performance on classification of retinal OCT scans. After training on historical clinical data, machine learning methods can offer the clinician support in the decision-making process. Care should be taken not to mistake neural network output as treatment recommendation and to ensure a final thorough evaluation by the treating physician.

Prediction of Local Response After Stereotactic Radiosurgery for Brain Metastases Using Convolutional Neural Networks based radiomics

A deep learning concept based on artificial convolutional neural networks (CNN) is regarded as an emerging radiomics methodology because it uses minimal amount of image preprocessing. The purpose of this study is to predict radiation response of metastatic brain tumor receiving stereotactic radiosurgery (SRS) using the radiomics model based on CNN. The BM lesions that received stereotactic radiosurgery (SRS) were reviewed and 110 tumors were selected for the study. On the planning CT images of the target tumors, a data-set was made by matching the single axial image which presents tumor center and the tumor label. The random assignment was repeated to generate 50 independent experimental combinations of data-sets. The same group is composed of 10 different data-sets combination methods, but the data-sets of the evaluation group are assigned equally. We trained the image of the training group and the matching label on the CNN we implemented. We used the validation data-sets to select the best-performing learning model. The results of this classification were compared to the matched label. CNN is a kind of artificial neural network in which the connectivity pattern between its neurons is inspired by the organization of visual cortex. The CNN of this study is implemented by TensorFlow which is a machine learning framework. The predicted AUC(area under the ROC curve) of individual CNN models for the 50 data-set combinations ranged from 0.602 [95% CI, 36.5% -83.9%] to a maximum of 0.826 [95% CI, 64.3% -100%]. -97.1%] to a maximum of 0.856 [95% CI, 68.2% - 100%]. CNN based ensemble radiomics models that learn BM 's SRS planning CT images and their early responses were able to predict SRS responses with high accuracy for unlearned BM images. This study showed for the first time the applicability of the CNN-based radiomics model to predict the prognosis of radiation therapy with small data.

Places: An Image Database for Deep Scene Understanding

The rise of multi-million-item dataset initiatives has enabled data-hungry machine learning algorithms to reach near-human semantic classification at tasks such as object and scene recognition. Here we describe the Places Database, a repository of 10 million scene photographs, labeled with scene semantic categories and attributes, comprising a quasi-exhaustive list of the types of environments encountered in the world. Using state of the art Convolutional Neural Networks, we provide impressive baseline performances at scene classification. With its high-coverage and high-diversity of exemplars, the Places Database offers an ecosystem to guide future progress on currently intractable visual recognition problems.

Tuning the Parameters of a Convolutional Artificial Neural Network by Using Covering Arrays

Tuning the Parameters of a Convolutional Artificial Neural Network by Using Covering Arrays

From retina to behavior: prey-predator recognition by convolutional neural networks and their modulation by classical conditioning

Visual object-recognition plays a crucial role in animals that utilize visual information. In this study, we address the prey-predator recognition problem by optimizing artificial convolutional neural networks, based on neuroethological studies on toads. After the optimization of the overall network by supervised learning, the network showed a reasonable performance, even though various types of image noise existed. Also, we modulated the network after the optimization process based on the computational theory of classical conditioning and the reinforcement learning algorithm for the adaptation to environmental changes. This adaptation was implemented by separated modules that implement the “innate” term and “acquired” term of outputs. The modulated network exhibited behaviors that were similar to those of real toads. The neural basis of the amphibian visual information processing and the behavioral modulation mechanism have been substantially studied by biologists. Recent advances in parallel distributed processing technologies may enable us to develop fully autonomous, adaptive artificial agents with high-dimensional input spaces through end-to-end training methodology.

Convolutional deep-learning artificial neural networks

This paper discusses the history of the appearance and development of the concept of convolutional artificial neural networks, which, because they use a learning technology based on back-propagation of the error signal, have become one of the most efficient tools of automatic image classification. Along with the possibilities of modern convolutional neural networks in the area of shape classification of objects, the features of their use in analyzing the information of other hierarchical levels have also been analyzed—from the classification of textures to the structural decomposition of images, based on the formation of attention zones. Convolutional networks are considered in close association with a description of their natural analogs, found in the neural ensembles of living visual systems.

On the Control of the Technical Condition of Elevator Ropes Based on Artificial Intelligence and Computer Vision Technology

Introduction. The safety problem and the situation with accidents during the operation of elevator installations are elucidated. The role of elevator rope defects as a factor of dangerous incidents is indicated from the point of view of statistics. The malfunctions of the elevator mechanical equipment related to the defective indices of the ropes are listed. There is a difference in the documentary fixation of defective indices and rejection rates of ropes of lifting structures. Materials and Methods. The well-known approaches to the control of ropes of lifting structures were described. It was emphasized that visual inspection control (VIC) was required to identify such rejection rates of steel elevator ropes as geometry change, corrosion and wear, wire breaks, temperature exposure, etc. The rejection rate was presented in the form of a mathematical system. The technical condition of elevator ropes during the operation was integrally assessed by the totality of identified defects at a fixed length. The decision to create a software and hardware complex (PAC) for the practical implementation of visual and measuring control was validated. Results. The developed PAC VIC laboratory sample consisted of a hardware part, a video stream processing module, communicator for the server connectivity, specially designed software, and a client mobile application. PAC VIC implemented the following functions: – automatic detection and classification of the major significant rope defects based on a deep convolutional artificial neural network; – demonstration of a three-dimensional image of a rope and an image scanning algorithm with distortion compensation, according to which the metric characteristics of defects were fixed; – integral assessment of the technical condition of the rope according to the totality of detected defects; – color interpretation of the actual technical condition of the rope with subsequent transmission to the user's mobile device. Preliminary tests have shown the suitability of the PAC VIC for identifying defects. The reliability of the results for the identification and qualification of defects exceeded 80%. Work on deep learning of the system continues. Discussion and Conclusions. PAC VIC of elevator ropes provides eliminating the risks of visual control caused by the psychophysical state of a person. It works remotely and contactless. The solution proposed by the authors automatically evaluates the rejection rates according to five criteria: external wire breaks, surface wear, rope diameter change, undulation, traces of temperature exposure. An important result of the VIC of steel ropes using computer vision and artificial intelligence is an increase in reliability and safety during the operation of elevator equipment.