Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
A Southwest Research Institute team has developed a machine learning tool to label large, complex datasets efficiently. The iterative labeling technique reduces manual verification time by 50%, enabling deep learning models to identify potentially hazardous solar events more accurately.
A UTSA professor will use a five-year $550,000 grant to study natural language processing and develop NLP models tailored to specific population groups. The goal is to improve the accuracy and relevance of these models in everyday applications.
Researchers have identified 42 genes related to 15 different cellular mechanisms that affect the risk of different types of somatic mutations. This comprehensive study may help explain cancer predisposition and potentially personalize prevention programs and cancer treatments.
A machine-learning model improved the simulation of European heatwave frequency by analyzing climate factors, with sea surface temperature contributing most to predictions. The model found that previous winter climate factors provided the best simulation results.
A cutting-edge AI system that mimics human gaze has been created by Cardiff University researchers. The system accurately predicts where humans are most likely to look in an image, with potential applications in robotics, multimedia communication, video surveillance, automated image editing, and medical diagnostics.
A new machine learning-based AI tool has been developed to aid doctors in distinguishing between tropical diseases such as dengue and malaria. The tool has shown promising results in improving diagnosis accuracy.
A team of researchers led by Danilo Vasconcellos Vargas has developed a new method called 'Raw Zero-Shot' to evaluate the robustness of artificial neural networks in image recognition. The study found that Capsule Networks produced the densest clusters, indicating improved transferability and potential solutions for improving AI robust...
A new algorithm forecasts crime with 90% accuracy, highlighting increased police response in wealthier areas and biased enforcement in disadvantaged neighborhoods. The tool divides cities into spatial tiles to detect patterns and predict future events.
A new robotic system, FuseBot, has been developed to efficiently retrieve buried objects in piles. The system uses radio frequency signals and computer vision to reason about the probable location and orientation of objects under the pile, enabling it to find more hidden items than a state-of-the-art robotics system in half the time.
Researchers used topological mathematics and machine learning to identify a hidden relationship between nano-scale structures and thermal conductivity in amorphous silicon. They found that the persistent homology diagram can be used as a descriptor for machine learning, achieving accurate predictions about thermal conductivities.
Researchers from Trinity College Dublin have proposed three guiding principles for improving AI, inspired by infant learning. The new approach aims to enhance machine learning with in-built preferences and developmental trajectories, using richer datasets that capture multiple senses.
A new AI system uses artificial neural networks to recognize objects more accurately and stably, despite changing visual inputs. The system mimics human eye movements to improve machine vision capabilities, reducing errors in self-driving cars and other applications.
A new AI-powered mental health application, FuturSelf, uses machine learning to identify the shortest path to mental stability. The system offers personalized recommendations for improving long-term well-being.
Researchers used wearable activity trackers and AI to identify indicative physiological changes in people with COVID-19, detecting infections up to 2 days before symptoms began. The study found that the algorithm was effective in identifying positive cases, but its accuracy needs further improvement.
A new approach based on deep learning AI detects weak gravitational signals, or PEGS, generated by large-mass motion in megaquakes. This allows for real-time tracking of earthquake growth after a magnitude 8 event.
A study published in JMIR Formative Research reveals that automated analysis of a common clinical drawing task can be used to estimate cognition in older adults from Japan and the USA. Researchers found that people with lower cognitive scores had more variability in their drawing speed, angle, and time spent paused during the task.
Researchers at UCLA, MIT and NIH found that AI systems can mimic human understanding of words by relating their meanings through co-occurrence. The study developed a technique called 'semantic projection' to analyze the system's knowledge and found it to be more intuitive than expected.
Researchers use machine learning to automatically analyze Reflection High-Energy Electron Diffraction (RHEED) data, enabling faster and more efficient discovery of new materials. The study focused on surface superstructures in thin-film silicon surfaces and identified optimal synthesis conditions using non-negative matrix factorization.
A team of researchers from SLAC National Accelerator Laboratory developed a mathematical technique using machine learning concepts to model a magnet's previous states. This approach eliminates the need to reset magnets and results in immediate improvements in accelerator performance.
Researchers have developed an AI-powered approach to calculate molecular spectra using Graph Neural Networks (GNNs), significantly reducing computation time and improving accuracy. The SchNet model achieved a 20% increase in accuracy while reducing computational time, enabling the analysis of complex molecules like quantum dots.
Researchers at Case Western Reserve University developed an AI tool that analyzes routine tissue biopsy images to predict response to immunotherapy in lung and gynecologic cancers. The study found that the AI consistently predicted clinical outcomes, including survival, for three common immunotherapies.
Researchers designed a modular AI chip that can be easily upgraded by swapping out layers, reducing the need for new devices. The chip uses optical communication to transmit information between layers, enabling high versatility in edge computing applications.
The study found that precipitation during cold seasons is a major contributor to energy consumption in Tibet, particularly when temperatures fall below freezing. Additionally, there is a marked disparity between urban and rural areas in how climate affects energy consumption, with urban areas experiencing higher rates on cold, snowy days.
A recent study published in Angewandte Chemie found that AI models struggle to predict reaction yields due to biased data, mainly caused by a lack of reported failed experiments. The researchers attribute this failure to three possible causes: experimental error, personal bias, and underreporting of negative results.
Researchers developed an AI model to analyze thyroid ultrasound images, identifying nodules unlikely to be cancerous with a sensitivity of 97%. This technology could reduce the number of unnecessary biopsies and assist radiologists in choosing which nodules should undergo biopsy.
A new machine learning algorithm called 'ikarus' has found a gene signature characteristic of tumors, distinguishing between healthy and tumor cells in various types of cancer. The algorithm was trained on single-cell sequencing data sets and demonstrated an extraordinarily high success rate in distinguishing between different cell types.
Researchers developed a machine learning model to predict NAFLD development based on gut microbiome data, showing 90% of subjects who developed the disease had subtle differences in their samples. The model combines easily measurable information from blood and microbiome data with high accuracy.
Researchers at MIT identified a flawed analysis of website-fingerprinting attacks and developed new techniques to prevent them. They found that attackers can use machine-learning algorithms to decode signals leaked between software programs, enabling them to obtain private information.
A study by UFZ researchers used satellite data and artificial intelligence to determine the land-use intensity of meadows and pastures in Germany. The results show that grassland was used less intensively in 2018 due to drought, with 64% not mowed compared to 36% in 2017.
Cong Shen earns CAREER award to develop 6G wireless communications systems supporting machine learning and AI technologies. His research aims to boost the performance of machine-learning models driving the Internet of Things while protecting data privacy.
A new study published in the Proceedings of the National Academy of Sciences offers fresh insights into infant word learning. Researchers found that infants between 7 and 11 months old learn words by building up memory representations over time, rather than through repeated connections between words and objects.
PaSh parallelizes Unix shell scripts automatically and accurately, boosting execution speeds by hundreds of times. Researchers tested the system on hundreds of scripts and found no errors, making it accessible to data scientists, biologists, engineers, and economists.
Researchers at MIT found that explanation methods used to aid human decision-makers in high-stakes situations often have lower accuracy for minoritized subgroups. The fidelity of these explanations varies dramatically between subgroups, with the quality often significantly lower for women and Black people.
A new AI-powered technique updates fuel inventories to better predict fire behavior and spread. The method, developed at NCAR, uses satellite imagery to account for pine beetle damage and was tested on the 2020 East Troublesome Fire in Colorado.
Researchers at the University of Rochester have created an automated scanning device that detects monolayers with high accuracy, reducing processing time and costs. The system utilizes AI-powered image processing to analyze images of materials, identifying monolayers with near 100% accuracy in just nine minutes.
A team of researchers from Waseda University developed a novel solution to efficiently solve complex optimization problems using Ising machines. Their hybrid algorithm reduces residual energy and reaches more optimal results in shorter time, increasing the machine's applicability across industries and sustainability practices.
A cohort study found that moderate coffee consumption, especially unsweetened coffee, is associated with a lower risk of death. The results suggest that adults who drink 1.5 to 3.5 cups of coffee per day may be less likely to die during a 7-year follow-up period compared to non-coffee drinkers.
Researchers at University of Tokyo used machine learning-based analysis of scalp-recorded EEG to see when and where odors are processed in the brain. Unpleasant odors were found to be processed earlier than pleasant ones, suggesting potential early warning system against dangers.
A study by University of Alberta researchers found that language related to depression on social media did not increase in response to COVID-19 waves, but rather decreased as people adapted. Instead, they found spikes in language related to the Black Lives Matter movement and other global events.
A new AI method can distinguish between the overall sounds of healthy and unhealthy coral reefs, making it a valuable tool for monitoring reef health. The technique uses machine learning to analyze sound recordings and track the progress of reef restoration projects.
Researchers use AI software to identify invasive species and native trees threatened by disease from photos taken by on-train cameras. This technology enables railway workers to take action to better manage lineside vegetation and achieve biodiversity net gain by 2035.
Researchers have found that mutational signatures, which reflect a collection of mutations across the genome, can accurately predict drug response in cancer cells. The study suggests that these signatures may hold the key to better cancer therapies and could be used to predict treatment response.
A Concordia study analyzed gender patterns in AI over two decades, finding women are making strides despite underrepresentation. Notably, female-male collaboration has increased, while core positions held by women remain scarce, tied to factors like family obligations and male-dominated environments.
A new ECG-based AI model developed by Tempus and Geisinger can predict patients at increased risk of undiagnosed structural heart disease. The rECHOmmend study found that the model achieved excellent performance, exceeding previous models in predicting any single disease.
A machine learning algorithm has outperformed astronomers in analyzing microlensing data to find new exoplanets, revealing connections hidden in complex mathematics from general relativity. The AI algorithm uncovered a degeneracy that had been missed by experts, suggesting a broader theory is likely incomplete.
Researchers have successfully processed sequences with a large neural network while consuming significantly less energy on neuromorphic hardware. This breakthrough showcases the potential of neuromorphic technology to improve the energy efficiency of AI workloads.
Researchers found that computer assistance in design leads to better solutions but compromises creativity and user agency. In a virtual reality experiment, novice designers outperformed their human-led counterparts when using an optimized approach.
Researchers at ETH Zurich have developed a novel memristor design that can switch between two operation modes, enabling greater efficiency in machine learning applications. This breakthrough component is made of halide perovskite nanocrystals and simulates complex neural networks with high accuracy.
A new distributed learning technique, GD-SEC, reduces communication requirements in wireless architecture, improving efficiency and reducing computational cost. The method employs data compression to transmit only meaningful, usable data, enhancing the impact of machine learning while minimizing its limitations.
The 27th North American Meeting will focus on technological challenges, breakthrough discoveries, and state-of-the-art research in catalysis. The meeting features plenary lectures by renowned experts in the field.
Researchers at Duke University have developed a machine learning algorithm that incorporates known physics into neural networks, allowing for new insights into material properties and more efficient predictions. The approach helps the algorithm attain transparency and accuracy, even with limited training data.
A new nanosensor platform uses machine learning to analyze spectral signatures of carbon nanotubes for early detection of ovarian cancer. The approach detects biomarkers and recognizes the cancer itself, offering a promising alternative to traditional methods.
Researcher Jundong Li aims to improve machine recommendations and predictions based on cause and effect. He develops a suite of algorithms and mathematical models informed by human experience and intuition to find cause-and-effect relationships in big data.
Researchers developed a data-driven robotic experiment system to identify electrolyte materials with desirable properties. They discovered a multi-component electrolyte that enhances the cycle life of lithium–air batteries, accelerating the development of next-generation rechargeable batteries.
A study by HBP scientists found that wakefulness, non-REM sleep, and REM sleep have complementary functions for learning: experiencing stimuli, solidifying experiences, and discovering semantic concepts. This research suggests that unusual dreams, simulated using Generative Adversarial Networks, can improve brain learning by introducin...
Xie aims to develop algorithms that reflect human domain knowledge and reasoning patterns, increasing trust in machine learning models. He also explores algorithmic fairness, considering multiple perspectives on what's considered fair, to address concerns about biased decision-making.
A new machine-learning framework has been developed to improve the design of catalysts, which speed up chemical reactions. The approach analyzes the conversion of carbon monoxide to methanol using a copper-based catalyst and identifies key steps that need to be tweaked to increase productivity.
An artificial intelligence model trained on histological images of surgical specimens accurately classified patients with and without Crohn disease recurrence. The model revealed previously unrecognized differences in adipose cells and mast cell infiltration, enabling stratification by prognosis for postoperative Crohn disease patients.
Researchers at Oak Ridge National Laboratory developed an algorithm that guides breakthroughs in new materials using self-driving microscopes. The approach combines physics and machine learning to automate microscopy experiments, enabling faster discoveries of functional materials' properties.
The university's new Robotics and Autonomous Systems Teaching and Innovation Center (RASTIC) will provide students with hands-on experience in robotics, autonomous systems, and self-driving technology. The lab aims to boost Massachusetts' competitiveness in the tech sector by supporting innovative projects and startups.