Within (semi-)automated visual industrial inspection, learning-based approaches for assessing visual defects, including deep neuralnetworks, enable the processing of otherwise small defect patterns in pixel size on h...
详细信息
ISBN:
(纸本)9798350318920;9798350318937
Within (semi-)automated visual industrial inspection, learning-based approaches for assessing visual defects, including deep neuralnetworks, enable the processing of otherwise small defect patterns in pixel size on high-resolution imagery. The emergence of these often rarely occurring defect patterns explains the general need for labeled data corpora. To alleviate this issue and advance the current state of the art in unsupervised visual inspection, this work proposes a DifferNet-based solution enhanced with attention modules: AttentDifferNet. It improves image-level detection and classification capabilities on three visual anomaly detection datasets for industrial inspection: InsPLAD-fault, MVTec AD, and Semiconductor Wafer. In comparison to the state of the art, AttentDifferNet achieves improved results, which are, in turn, highlighted throughout our quali-quantitative study. Our quantitative evaluation shows an average improvement compared to DifferNet - of 1.77 +/- 0.25 percentage points in overall AUROC considering all three datasets, reaching SOTA results in InsPLAD-fault, an industrial inspection in-the-wild dataset. As our variants to AttentDifferNet show great prospects in the context of currently investigated approaches, a baseline is formulated, emphasizing the importance of attention for industrial anomaly detection both in the wild and in controlled environments.
Activation functions play a pivotal role in determining the training dynamics and neural network performance. The widely adopted activation function ReLU despite being simple and effective has few disadvantages includ...
详细信息
ISBN:
(纸本)9781665493468
Activation functions play a pivotal role in determining the training dynamics and neural network performance. The widely adopted activation function ReLU despite being simple and effective has few disadvantages including the Dying ReLU problem. In order to tackle such problems, we propose a novel activation function called Serf which is self-regularized and non-monotonic in nature. Like Mish, Serf also belongs to the Swish family of functions. Based on several experiments on computer vision (image classification and object detection) and natural language processing (machine translation, sentiment classification and multi-modal entailment) tasks with different state-of-the-art architectures, it is observed that Serf vastly outperforms ReLU (baseline) and other activation functions including both Swish and Mish, with a markedly bigger margin on deeper architectures. Ablation studies further demonstrate that Serf based architectures perform better than those of Swish and Mish in varying scenarios, validating the effectiveness and compatibility of Serf with varying depth, complexity, optimizers, learning rates, batch sizes, initializers and dropout rates. Finally, we investigate the mathematical relation between Swish and Serf, thereby showing the impact of pre-conditioner function ingrained in the first derivative of Serf which provides a regularization effect making gradients smoother and optimization faster.
The performance of Convolutional neuralnetworks (CNNs) is critically dependent on the underlying computational configurations, particularly in terms of processing capabilities and data handling techniques. As deep le...
详细信息
Artificial reasoning systems via Artificial Intelligence (AI) and machinelearning (ML) have made tremendous progress within the past decade. AI/ML systems have been able to reach unprecedented new levels of autonomy ...
详细信息
ISBN:
(纸本)9781510674219;9781510674202
Artificial reasoning systems via Artificial Intelligence (AI) and machinelearning (ML) have made tremendous progress within the past decade. AI/ML systems have been able to reach unprecedented new levels of autonomy for a multitude of applications ranging from autonomous vehicles to biomedical imaging. This new level of intelligence and freedom for AI/ML systems requires them to have a degree of human-like intelligence in terms of causation beyond the correlation. This, however, has remained a major challenge for investigators when combining causality with AI/ML systems. AI/ML systems that are capable of generating cause and effect relationships are still in their infancy, as the literature highlights. The lack of investigations for causal reasoning systems that are capable of using datasets other than tabular data is well highlighted within literature. Causal learning for image, audio, video, radio-frequency, and other modalities still remain a major challenge. While there are open-source tools available for causal learning with tabular data, there is a lack of tools for other modalities. To this extent, this study proposes a causal learning method with image datasets by using existing tools and methodologies. Specifically, we propose to use existing causal discovery toolboxes for investigating causal relations within image datasets by converting image datasets into tabular form with feature extraction using tools such as auto-encoders and deep neuralnetworks. The converted dataset can then be used to generate causal graphs by using tools such as the Causal Discovery Toolbox to highlight the specific cause and effect relations within the data. For AI/ML systems using causal learning for image datasets via existing tools and methodologies can provide an extra layer of robustness to ensure fairness and trustworthiness.
In the recent years, Spiking neuralnetworks have gain much attention from the research community. They can now be trained using the powerful gradient descent and have drifted from the neuroscience to the machine Lear...
详细信息
ISBN:
(纸本)9781665488679
In the recent years, Spiking neuralnetworks have gain much attention from the research community. They can now be trained using the powerful gradient descent and have drifted from the neuroscience to the machinelearning community. An abundant literature shows that they can perform well on classical Artificial Intelligence tasks such as image or signal classification while consuming less energy than state-of-the-art models like Convolutional neuralnetworks. Yet, there is very little work about their performance on unsupervised anomaly detection and time-series prediction. Indeed, the processing of such temporal data requires different encoding and decoding mechanisms and rises questions about their capacity to model a dynamical signal with long term temporal dependencies. In this paper, we propose for the first time a Sparse Recurrent Spiking neural Network with specific encoding and decoding mechanisms to successfully predict time-series and do Unsupervised Anomaly Detection. We also provide a framework to describe in detail our model computational costs and fairly compare them with state-of-the-art models. Despite improvable performances, we show that our model perform well on these tasks and open a door for further studies of such applications for Spiking neuralnetworks.
ActivMedica is an innovator in brain tumor diagnosis in the rapidly changing healthcare industry, employing cutting edge AI tools including deep learning and natural language processing (NLP). The shortcomings of exis...
详细信息
Breast cancer (BC) is a leading cause of cancer-related deaths in women, but early detection significantly improves survival rates. Recently, deep learningneuralnetworks have shown potential for enhancing BC screeni...
详细信息
This research focuses on generating image captions using Convolutional neuralnetworks (CNN) and Long Short-Term Memory (LSTM) models. As deep learning advances, the availability of large datasets and increased comput...
详细信息
Analysis of medical image is new technique to solve the medical problems by examine the images that has been generated through some detailed clinical practice. For better clinical diagnosis, the goal is to extract inf...
详细信息
ISBN:
(纸本)9789819751990
Analysis of medical image is new technique to solve the medical problems by examine the images that has been generated through some detailed clinical practice. For better clinical diagnosis, the goal is to extract information in a way that is both affective and efficient. Abdominal bleeding, ulcers, tumors, celiac disease, ulcerative colitis, and other gastrointestinal disorders are difficult to diagnose due to the uncertainty of accessing a volute condition in the human body. WCE is a non-invasive, painless, and patient-controlled gastrointestinal tract examination. Although using deep learning models to detect problems in WCE photos enhances detection accuracy, model training necessitates a large volume of labeled data. These deep models, on the other hand, are difficult to describe and do not incorporate expert input into the model decision-making process. Deep learning-based machinelearning breakthroughs have improved the ability to categories, detect, and quantify patterns in medical images. Deep learning approaches are quickly being applied in medical imaging to increase performance in a variety of medical applications. Due to recent advancements in the field of biomedical engineering, medical image analysis has become one of the most important research and development areas. The use of machinelearning algorithms for the processing of medical images is one of the reasons for this progress. Deep learning has been successfully applied as a machinelearning technology, allowing a neural network to learn information automatically. This contrasts with methods that rely on traditional handcrafted elements. The selection and computation of these characteristics are a difficult task. Deep convolution networks, a type of deep learning approach, are widely employed in medical imageprocessing. With these considerations in mind, the purpose of this research is to see if employing semi-supervised deep learning models over supervised deep learning methods for detecting a
Deep neuralnetworks (DNNs) exhibit superior performance in various machinelearning tasks, e.g., image classification, speech recognition, biometric recognition, object detection, etc. However, it is essential to ana...
详细信息
ISBN:
(纸本)9798350370287;9798350370713
Deep neuralnetworks (DNNs) exhibit superior performance in various machinelearning tasks, e.g., image classification, speech recognition, biometric recognition, object detection, etc. However, it is essential to analyze their sensitivity to parameter perturbations before deploying them in real-world applications. In this work, we assess the sensitivity of DNNs against perturbations to their weight and bias parameters. The sensitivity analysis involves three DNN architectures (VGG, ResNet, and DenseNet), three types of parameter perturbations (Gaussian noise, weight zeroing, and weight scaling), and two settings (entire network and layer-wise). We perform experiments in the context of iris presentation attack detection and evaluate on two publicly available datasets: LivDet-Iris-2017 and LivDet-Iris-2020. Based on the sensitivity analysis, we propose improved models simply by perturbing parameters of the network without undergoing training. We further combine these perturbed models at the score-level and at the parameter-level to improve the performance over the original model. The ensemble at the parameter-level shows an average improvement of 43.58% on the LivDet-Iris-2017 dataset and 9.25% on the LivDet-Iris-2020 dataset. The source code is available at https://***/redwankarimsony/WeightPerturbation-MSU.
暂无评论