Advanced searches left 3/3

Artificial Intelligence - Astrophysics Data System

Summarized by Plex Scholar
Last Updated: 10 September 2022

* If you want to update the article please login/register

Explainable Artificial Intelligence to Detect Image Spam Using Convolutional Neural Network

Image spam security analysis has long been a common field of study, particularly due to the internet's explosive growth. We use CNN model to categorize image spam in this work, while real XAI tools such as Local Interpretable Model Agnostic Explanation and Shapley Additive Explanations were used to provide explanations for the decisions made by the black-box CNN models regarding spam image detection's decision. We collect spam images and normal images from three different publicly available email corporas and then analyze the results of the new program on a 6636 image database.

Source link: https://ui.adsabs.harvard.edu/abs/2022arXiv220903166Z/abstract


Megathrust Seismicity Through the Lens of Explainable Artificial Intelligence

These three types of nodes in a Fully Connected Network FCN's input layer are used to determine earthquake magnitude embedding the state of the region b, the transmission device c, and the subsequent seismicity a We then analyzed the trained network to determine the relative weights of the input nodes, providing vital information on the mechanisms that govern seismicity in a region, their size, and likelihood.

Source link: https://ui.adsabs.harvard.edu/abs/2022EGUGA..2411060G/abstract


Can landslide inventories developed by artificial intelligence substitute manually delineated inventories for landslide hazard and risk studies?

Landslide investigators are required for landslide susceptibility mapping, hazard modeling, and further risk mitigation efforts. For decades, researchers and companies around the world have favored manual visual interpretation of satellite and aerial photos. The automatic generation of landslide inventories utilizing Artificial Intelligence methods is still in its early stages; as currently there is no published study that can create a ground truth representation of landslide conditions following a landslide triggering event, a landslide generator is still in its early stages. Using AI-based methods, evaluation results in recent literature show a range of 50-80% of F1-score in terms of landslide boundary delineation using AI-based methods. However, only few studies claim to have received more than 80% F1 score, with the exception of those using the testing of their model evaluation in the same research area. There is also a research gap between the generation of AI-based landslide generators and their usability for landslide hazard and risk studies.

Source link: https://ui.adsabs.harvard.edu/abs/2022EGUGA..2411422M/abstract


From data to noise to data for mixing physics across temperatures with generative artificial intelligence

Temperatures at which simulations were never carried out are never carried out here, we show how to blend data from simulations conducted at a set of temperatures and create molecular models at any temperature of concern, including temperatures at which simulations were never carried out. The configurations we produce have accurate Boltzmann weights, and our algorithm minimizes the creation of spurious unphysical configurations.

Source link: https://ui.adsabs.harvard.edu/abs/2022PNAS..11903656W/abstract


Identifying conditions that sculpted bedforms - Human insights to build an effective artificial intelligence 'AI'

Human intelligence data, or 'AI', can be used to help design an effective machine learning algorithm, as shown by insights from a geoscience communication campaign, which was confirmed using preliminary tests with an artificial neural network.

Source link: https://ui.adsabs.harvard.edu/abs/2022EGUGA..24.1437H/abstract


Camera Rain Gauge Based on Artificial Intelligence

Because of their costs and installation difficulties, flood risk monitoring, alert, and adaptation in urban areas need near-real-time fine-scale precipitation measurements that are impossible to obtain from currently available measurement networks due to their costs and installation difficulties. This review describes unprecedented system for rainfall monitoring based on artificial intelligence, particularly the use of deep learning for computer vision in camera photographs. With Convolutional Neural Networks, rainfall is measured directly from single photographs to Deep Learning models based on transfer learning. The prototype was used in a real-world operational environment using a pre-existing 5G surveillance camera. Lastly, a case study over the Matera urban area was used to illustrate the dangers and limitations of rainfall surveillance using camera-based sensors. The binary classifier's results demonstrated high stability and portability: the test and deployment algorithms suffered from drastic accuracy changes, according to the image source. Scenes that are also confusing for human visual perception are not suitable for this method. Where remote-sense rainfall data is missing or has broad agreement in connection with the scale of the study, remote-sensed rainfall data is inaccessible or has broad resolution respect with the scope of the study. Future research will continue to refine precision and increase availability from a smart city perspective, with incremental learning algorithms and greater data collection via experiments and crowdsourcing.

Source link: https://ui.adsabs.harvard.edu/abs/2022EGUGA..24.4266A/abstract


Assessment of the variation of soil trace metals using artificial intelligence: A case study from Eastern Province, Saudi Arabia

Metals from many sources that might pollute the atmosphere may be preserved by Soils. In selected areas in Eastern Province, Saudi Arabia, this research involved the spatial evaluation of topsoil contamination with trace metals. The samples were analyzed for various trace metals by inductively coupled plasma spectroscopy. Model results showed that agricultural and industrial stations' performance was highly regarded with goodness-of-fit ranges of 51-91% and 99%, respectively. This report revealed that AI models could be used to quickly assess soil trace metals and a decision support system.

Source link: https://ui.adsabs.harvard.edu/abs/2022EGUGA..24.8261Y/abstract

* Please keep in mind that all text is summarized by machine, we do not bear any responsibility, and you should always check original source before taking any actions

* Please keep in mind that all text is summarized by machine, we do not bear any responsibility, and you should always check original source before taking any actions