Articles tagged with Deep Learning
Deep Nanometry enables high-speed analysis of nanoparticles, detecting even trace amounts of rare particles like extracellular vesicles indicative of colon cancer. This technique has potential applications in various fields including vaccine development and environmental monitoring.
Researchers developed an innovative model, ECG-LM, that leverages large language models to interpret complex ECG signals. The model improves the accuracy and speed of heart-related diagnostics, particularly in resource-limited environments.
A team of researchers from Sophia University demonstrated the potential of combining social media posts and transformer-based learning models to detect heat stroke risks. The study found that a Japanese model called LUKE achieved high accuracy, suggesting its viability in monitoring heat stroke risks during heat waves.
Researchers have created a geometric deep learning approach called MARBLE that can infer latent brain activity patterns across experimental subjects. The method uses dynamic motifs to analyze neural population recordings and has been shown to be more interpretable than other machine learning methods.
Researchers used deep learning models to compare gene regulation in different cell types of human and chicken brains, shedding new light on brain evolution and providing tools for studying gene regulation. The study found that while some cell types are highly conserved between birds and mammals, others have evolved differently.
A new deep learning technique called CTLESS enhances myocardial perfusion imaging accuracy without requiring additional radiation scans. This method leverages deep learning to estimate attenuation maps, improving diagnostic interpretation and potentially boosting technological health equality across the U.S. and worldwide.
Researchers develop a new framework called MSF-Net to improve WiFi-based human activity recognition, achieving high accuracy scores compared to state-of-the-art techniques. The technology has potential applications in smart homes, rehabilitation medicine, and care for the elderly.
Researchers have proposed a new fault-tolerant framework for distributed deep learning model training that minimizes overhead and improves efficiency. By utilizing idle system resources during training, the framework effectively coordinates tasks with fault-tolerance functions.
A recent study introduces an innovative method for analyzing body composition, providing accurate assessments of body fat and muscle distribution. The approach utilizes deep, nonlinear methods to enhance estimation accuracy, surpassing previous linear models.
A recent study by University of New Hampshire researchers explores how to standardize rock climbing route difficulty through machine learning techniques, aiming to promote inclusivity and accuracy. The most successful approach used route-centric natural language processing methods, achieving an accuracy of 84.7%.
A new deep learning model has been developed to detect and segment lung tumors on CT scans with high accuracy. The model achieved 92% sensitivity and 82% specificity in detecting lung tumors, outperforming physician-delineated volumes in some cases. However, further research is needed to improve the model's performance for larger tumors.
Researchers at Sharjah University have designed techniques to automatically predict suitable image dimensions using deep learning models. The proposed methods aim to bridge the gap in automating image retargeting approaches based on an image and target resolution.
The EPFL team has developed a deep-learning pipeline called MaSIF to design new proteins that interact with therapeutic targets. They have successfully designed novel protein binders that can recognize and bind to drug-protein complexes, offering potential applications in cell-based therapies and biosensors.
A new 6D pose dataset has been introduced, providing high-quality RGB and depth images with annotated 6D pose data. This dataset achieves state-of-the-art accuracy rates of 97.05% and 98.09% for robotic grasping and automation applications.
Researchers developed an AI-powered technology that transforms low-resolution, label-free images into high-resolution, virtually stained ones without fluorescent dyes. This innovation delivers stable and accurate cell visualization, overcoming limitations of traditional imaging methods.
Researchers at the University of Washington have developed new proteins that can neutralize lethal snake venom toxins using deep learning computational methods. These protein designs show promise for creating safer and more cost-effective antivenoms, potentially saving millions of lives annually.
Researchers have developed a new geometric machine learning method called MaSIF, which enables the design of proteins that bind specifically to desired molecular structures. This approach accelerates precision drug development by allowing for precise dosing and control of biological drugs.
Researchers developed an explainable deep learning model to predict and analyze HABs in Chinese lakes and reservoirs, achieving significant improvement over conventional machine learning methods. The model identified water temperature as the most influential factor driving algal bloom dynamics.
A novel method called RESQUE predicts computational and energy costs for updating deep learning/AI models, allowing users to make informed decisions about when to update models to improve AI sustainability. The researchers conducted extensive experiments to validate the performance of RESQUE.
Researchers developed TLE-PINN to predict melt pool morphology in selective laser melting, achieving superior accuracy and faster training times. The framework combines physics-informed constraints with deep learning techniques, enabling precise and efficient solutions for real-time process control and manufacturing optimization.
The one-core-neuron system (OCNS) minimizes model size while maintaining high performance in deep learning. OCNS employs a single neuron to encode high-dimensional data into a one-dimensional time-series representation, reducing parameters by an average of 0.035%.
The SSIT model uses a single encoder to extract spatial features and a decoder to reconstruct images with desired content and style. It outperforms other GAN models in image transformation tasks, offering potential for democratizing image transformation on devices like smartphones.
Researchers developed a deep learning model that classifies pancreatic cancer into molecular subtypes using histopathology images, achieving high accuracy and rapid turnaround time. The AI tool has the potential to improve patient outcomes by enabling timely and tailored treatment strategies.
A deep learning model developed by researchers at San Diego State University can accurately diagnose chronic obstructive pulmonary disease (COPD) using a single inhalation lung CT scan. The study found that the model performed similarly to traditional two-phase CT measurements, with added clinical data improving accuracy.
Researchers have developed an algorithm-based scheme to help drivers avert drowsiness, which contributes to thousands of fatal incidents and injuries every year. The system uses EEG signal detection and machine learning algorithms to achieve high accuracy and reduce training time.
A new computational model called Multi-Stage Residual-BCR Net (m-rBCR) uses a unique frequency representation to solve deconvolution tasks with fewer parameters and faster processing times. The model demonstrates high performance on various microscopy datasets, outperforming traditional methods.
Researchers have developed a deep-learning-powered metalens imaging system that overcomes limitations of traditional metalenses. The system pairs a mass-produced metalens with an image restoration framework driven by AI to achieve aberration-free, full-color images while maintaining compact form factor.
Researchers developed a deep learning-based method for identifying 2D materials using Raman spectroscopy, achieving high classification accuracy and reducing manual intervention. The new approach generates synthetic data to enhance datasets, enabling precise material characterization even with scarce experimental data.
A deep learning AI model can identify pathology in images of animal and human tissue much faster and often more accurately than people, potentially revolutionizing disease-related research and medical diagnosis. The model was trained using images from past epigenetic studies and showed accuracy comparable to human experts.
The study explores gene fusion technologies, including FISH, PCR, IHC, ECL, and NGS, to detect biomarkers in tumor diagnosis. AI-driven detection and comprehensive genome-wide analysis using NGS and bioinformatic tools enhance diagnostic accuracy.
Researchers introduce a novel approach to multiplexed fringe projection profilometry using deep learning and frequency-domain multiplexing. This method achieves high-resolution and high-speed 3D imaging at near-one-order of magnitude-higher frame rates with conventional low-speed cameras.
Researchers developed a novel noninvasive choroidal angiography method using deep learning, enabling layer-wise visualization and evaluation of choroidal vessels. The approach employs an advanced segmentation model to handle varying quality of OCT B-scans, offering a promising tool for clinical applications.
A new study from Virginia Tech shows large language models can assess human-made environments using street-view images, similar to traditional methods. LLM-based performance offers a more accessible tool for users in small to medium-sized cities, making it easier to manage smart urban infrastructure.
Researchers are using deep learning to help protect chimpanzees in the Greater Mahale Ecosystem, Tanzania. A new acoustic detector has been developed to identify chimpanzee sounds and monitor population density more efficiently, allowing for better conservation strategies.
A new review article explores the transformative role of deep learning techniques in revolutionizing protein structure prediction. Deep learning models like AlphaFold 2 have achieved high accuracy, over 98%, in predicting human protein structures, surpassing traditional methods.
A novel collaborative framework integrates semi-supervised learning techniques to improve MRI segmentation accuracy, even with limited labeled data. The approach achieves high Dice scores and demonstrates its potential for practical clinical application.
A new study finds that AI-powered models exhibit similar levels of accuracy as ophthalmologists in identifying infectious keratitis, a leading cause of corneal blindness worldwide. The AI models displayed a sensitivity and specificity of 89.2% and 93.2%, respectively, matching the diagnostic accuracy of human experts.
Deep learning models used in remote sensing tasks are susceptible to various types of noise and attacks, compromising their performance. The study assesses the vulnerabilities of DL algorithms for object detection, revealing several weaknesses that can be leveraged by attackers.
Researchers at Chung-Ang University developed a novel GAN model, PMF-GAN, to address stability and efficiency issues. The model utilizes kernel functions and histogram transformations to improve the generator's ability to produce diverse outputs, reducing mode collapse and gradient vanishing.
A new study uses deep learning to infer the frequency of atmospheric blocking events over the past 1,000 years, shedding light on their potential impact under climate change. The model was trained using historical data and large ensembles of climate model simulations.
Researchers used deep learning to correlate citizen science data with remote sensing images, predicting plant distributions down to scales of a few square meters. The AI model, Deepbiosphere, outperformed previous methods in accuracy and showed potential for global monitoring of vegetation change.
A new study published in Scientific Reports reveals the importance of foot movement in early infant development and interaction. By using machine and deep learning techniques, researchers found that AI can accurately classify five-second clips of 3D infant movements, with foot movements showing the highest accuracy rates.
A team developed an AI system to analyze label-free photoacoustic histological images of human liver cancer tissues, achieving 98% accuracy in distinguishing between cancerous and non-cancerous cells. The integration of PAH with AI reduces tissue biopsy time and enhances reliability.
Researchers have developed an AI technology that can analyze mammary tissue biopsies to identify signs of damaged cells, a key indicator of breast cancer risk. The study found the AI was far better at predicting risk than current clinical benchmarks, offering improved treatment options for women.
A team of OU scientists, led by Nathan Snook, will use deep learning techniques to analyze numerical simulations of tornadoes. The goal is to improve tornado forecasting by identifying key factors that influence their formation.
Researchers from the University of Toronto's Rotman School of Management found that campaign size, social capital, and reward options are top factors in success. Machine learning identified a sweet spot for campaign duration and reward options, with success plateauing after 50 options.
A Concordia-led team developed a framework that enables crowdsourced deep reinforcement learning as a service, using blockchain technology. This allows smaller organizations to access complex AI tasks previously out of reach, reducing costs and risk.
Researchers introduced a novel illumination beam design based on deep learning, eliminating the need for sophisticated optics tools. The approach enhances image quality by optimizing both the deep learning network and the illumination beam simultaneously.
A novel deep learning model, DS-ViT-ESA, was developed to predict lithium battery lifespan with high accuracy using only a small amount of charging cycle data. The model achieved low prediction errors even when tested on unseen charging strategies, demonstrating its zero-shot generalization capability.
Researchers leverage deep learning networks to recover and enhance compromised metrics in biophotonic image data. This approach improves imaging speed and quality, allowing for high-fidelity all-in-focus images and efficient reconstruction with reduced data acquisition.
A breakthrough technology allows for touchless infrared imaging to monitor changes in pupil size and gaze direction behind closed eyes. This innovation can help identify wakefulness, awareness, and pain in sleep, anesthesia, and intensive care, enabling more accurate clinical decision-making.
A novel approach to overcome limitations of traditional methods, NeuPh uses local conditional neural fields to reconstruct high-resolution phase information from low-resolution measurements. It provides robust resolution enhancement and outperforms existing models in accuracy.