Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
A study by Cameron Buckner using deep neural networks suggests that human knowledge stems from sensory experience, a school of thought known as empiricism. The networks demonstrate how abstract knowledge is acquired and can be used to understand complex tasks in neuroscience and psychology.
Researchers from MIT CSAIL developed a system that uses machine learning to determine if a news source is accurate or biased. The system achieved 65% accuracy in detecting factuality levels and 70% accuracy in detecting bias, using common linguistic features across the source's stories.
A study using machine learning and data from 550 children with varying learning difficulties identified four clusters of cognitive challenges that did not match previous diagnoses. The researchers found that working memory and phonological skills deficits often overlap, affecting both reading and math abilities.
Scientists from the Salk Institute and UC San Diego use reinforcement learning to train gliders to navigate atmospheric thermals, reaching heights of 700 meters. The research highlights the role of vertical wind accelerations and roll-wise torques as navigational cues for soaring birds.
Researchers from UC San Diego use reinforcement learning to train gliders to navigate atmospheric thermals, achieving heights of 700 meters. The study highlights vertical wind accelerations and roll-wise torques as key cues for soaring birds, with implications for autonomous flying vehicle development.
New research reveals infants can acquire object categories using just a few labeled examples, sparking the process of categorization. This 'semi-supervised learning' strategy efficiently integrates all subsequent objects into their evolving category representation.
The MIT-developed AI model can associate specific words with specific patches of pixels in an image, enabling real-time object highlighting based on spoken descriptions. This innovation holds promise for applications such as language translation and automatic image annotation.
A new AI program can analyze lung tumor images to specify cancer types and identify genetic changes, offering a potential solution for faster diagnosis and treatment. The tool achieved high accuracy rates in distinguishing between adenocarcinoma and squamous cell carcinoma, as well as identifying altered genes linked to lung cancer.
Researchers from UPNA won a second-place award at a scientific meeting for their work on modifying the machine learning process with Choquet integral, improving deep neural network performance in classification problems. The collaboration involved researchers from Brazil and institutions in Navarre.
Using deep learning algorithms, researchers have developed a system that forecasts aftershocks significantly better than random assignment. By analyzing earthquake data and physics-based models, they identified the second invariant of the deviatoric stress tensor as an important factor in predicting aftershock locations.
Researchers developed an AI system to detect tiny lung cancer tumors in CT scans, achieving 95% accuracy compared to human eyes at 65%. The system uses a connection-based approach similar to facial recognition software to identify patterns.
Xiaochen Guo, a Lehigh University professor, aims to improve data movement efficiency by revamping memory systems. Her goal is to proactively create and redefine locality in hardware, unlocking fundamental improvements for machine learning applications.
Researchers trained a machine learning algorithm to analyze microscopic radiation damage, achieving an accuracy of 86% compared to humans. The algorithm can process images faster and more efficiently than humans, making it a promising tool for developing safe nuclear materials.
Researchers at the University of Alberta developed a machine-learning algorithm to analyze brain function images and predict schizophrenia diagnosis with 78% accuracy. The model also identified patients who would respond positively to antipsychotic treatment risperidone with 82% accuracy.
Researchers developed a technique to quickly teach robots novel traversal behaviors with minimal human oversight, enabling autonomous navigation in environments. The goal is to provide reliable robot teammates to the Soldier, allowing for faster task completion and enhanced situational awareness.
Scientists developed a new machine learning method that can make accurate predictions for all possible conditions governed by the same physical dynamics. The method provides simple and interpretable descriptions of underlying physics, allowing for safer robot operation.
The researchers developed a novel synaptic architecture that could lead to a new class of information processing systems inspired by the brain. Prototype chips containing over one million nanoscale memristive devices were used to implement a neural network for detecting hidden patterns and correlations in time-varying signals.
A new approach uses machine learning to generate computer-generated X-rays to supplement real images, increasing the size of training sets for AI systems. This method improves classification accuracy for common and rare conditions by up to 40%, overcoming a challenge in applying artificial intelligence to medicine.
Researchers developed a deep-learning model to predict biological age of muscles and estimate the importance of genetic and epigenetic factors driving muscle aging. The study identified tissue-specific biomarkers of aging, which can be used to track the effectiveness of interventions.
A new type of personalized machine learning helps robots accurately assess children's engagement and interest during autism therapy, with a correlation score of 60% compared to human experts. This technology aims to augment human therapists with key information to personalize therapy content and create more engaging interactions.
Three researchers at Argonne National Laboratory, Prasanna Balaprakash, Karen Mulfort, and Zhang Jiang, have earned the DOE's Early Career Research Program awards. They will receive funding to advance their research in machine learning, molecular interactions, and advanced materials imaging.
PlinyCompute is a system designed for developing high-performance big data codes, offering faster implementation of complex object manipulation and library-style computations compared to Spark. The platform was developed by Rice University's DARPA-funded Pliny Project team, which aims to create sophisticated programming tools using mac...
Scientists developed a machine learning approach to predict microbial pathways, allowing for faster design and development of biofuels. The method accurately predicted biofuel production profiles, outperforming traditional kinetic models.
Researchers at the University of Sydney have developed a generalized method to predict epileptic seizures using data from non-surgical devices powered by AI and machine learning. The system can alert epilepsy sufferers within 30 minutes of the likelihood of a seizure, with an accuracy rate of up to 81.4%.
Researchers have developed an automated atom fabrication process using machine learning, paving the way for mass production of atom-scale devices. This breakthrough aims to reduce energy consumption by 1000 times and increase computation speed a hundredfold, making it a game-changing technology for the information age.
Researchers developed a brain-computer interface that uses mutual learning to improve control for tetraplegic individuals. The system allowed users to adapt and learn at their own pace, leading to significant improvements in performance.
Researchers at University of Lincoln develop machine learning algorithms for self-learning robots in hazardous nuclear sites, increasing capabilities in waste handling and site monitoring. The project aims to build systems that can adapt to unique radiation conditions using vision-guided robot grasping, manipulation, and cutting.
Researchers at Johns Hopkins Medicine have made significant discoveries about the cerebellum's role in learning and prediction. By studying monkey brains, they found that Purkinje cells communicate through simple spikes (predictions) and complex spikes (error feedback), organizing into small groups to learn together.
A new study from Columbia University uses an AI algorithm to analyze the full range of behaviors exhibited by the tiny pond-dwelling creature Hydra. By comparing its movements to neural firing patterns, researchers hope to gain insights into how its nervous system works.
Researchers at the U.S. Army Research Laboratory developed a new technology that allows soldiers to learn 13 times faster than conventional methods, potentially saving lives. This technique uses low-cost hardware and collaborative filtering to achieve a significant speedup in training compared to state-of-the-art systems.
Researchers at UCSB are developing a chip that uses ionic memristor technology to create a physically unclonable device, rendering it vulnerable to cyber attacks. The technology aims to prevent cloning and hijacking of devices in networks, making them ideal for securing IoT devices.
Researchers used a deep learning algorithm to classify real galaxies in Hubble images, achieving remarkable consistency in its classifications. The study found that the algorithm identified a specific mass range for the 'blue nugget' phase of galaxy evolution, which is followed by quenching of star formation in the central region.
Researchers will use naturalistic approach with video vignettes and head-mounted cameras to measure infant eye direction and analyze their visual attention. The study aims to understand how young learners generate data for optimal language learning.
The University of Surrey has developed a new algorithm that can compress large amounts of data from bridge monitoring systems, reducing the storage requirements for authorities. The K-SVD method achieves near-lossless reconstruction with less than 0.1% data loss, compared to other methods which require up to 50% of the original data.
A team of scientists has developed a machine learning algorithm that can quickly identify new blends of ingredients for metallic glass, accelerating the discovery process by 200 times. The method uses data from thousands of experiments to pinpoint potential materials and has significant implications for the future of materials science.
A team of scientists has developed a method to discover new metallic glass alternatives using machine learning and accelerated experiments, reducing the discovery time from decades to hours. The approach enables researchers to quickly narrow down potential materials and get immediate feedback from AI models.
A team of scientists, including those from Google, developed a computer program that can identify structures in unstained brain cells. The program learned to spot features such as cell nuclei, dead cells, and specific types of brain cells by analyzing stained images.
A groundbreaking study by Gladstone and Google AI uses deep learning to analyze cell images, identifying features that humans can't detect. The method uncovers important information that was previously impossible or problematic for scientists to obtain.
Researchers used machine learning to classify abnormal protein activity in tumors, identifying potential 'hidden responders' who may benefit from specific therapies. The study combined genetic data with machine learning approaches to predict response to inhibitors affecting cancer cells with overactive Ras signaling.
A bioinformatics professor and his colleague used GitHub as their writing platform to collaborate on a paper about deep learning in biology and medicine. The paper has been massively rewritten and revised by online collaborators, resulting in over 40 co-authors.
Scientists at Bar-Ilan University discovered that learning in the brain occurs in dendrites, not just synapses. This new finding suggests a faster and more efficient learning process, with implications for current treatments and machine learning algorithms.
Researchers at Massachusetts General Hospital developed an artificial intelligence technique, AUTOMAP, that enables the production of high-quality images in less time and with lower doses. This approach uses deep learning to automatically determine the correct image reconstruction algorithm, allowing for instant feedback during scanning.
Researchers developed a visible neural network, DCell, that uses real-world cellular behaviors and constraints to predict cellular growth. The system can simulate cellular growth nearly as accurately as a real cell grown in a laboratory.
The UTSA researchers have created a cloud-based learning platform for artificial intelligence that aims to teach machines to learn like humans. This platform can help AI agents automatically detect threats in network traffic and improve their performance in discovering and thwarting new attacks.
Machine learning techniques can reconstruct a quantum system based on relatively few experimental measurements, allowing scientists to thoroughly probe complex systems exponentially faster than conventional methods. This method benefits the development of quantum computers and other applications of quantum mechanics.
Researchers developed an AI-based computational tool to screen patients for common retinal diseases, achieving over 95% accuracy in diagnosis. The tool also showed promise in diagnosing childhood pneumonia with over 90% accuracy, highlighting its potential applications in healthcare.
Researchers create a new approach to machine learning using a single-layer neural network that can analyze images with limited training data. The algorithm, called MS-D, requires far fewer parameters than traditional methods and has the ability to learn from a remarkably small set of images.
Computer scientists and materials researchers developed a more accurate and objective method for classifying steel microstructures. The method uses machine learning to analyze microscope images and achieve accuracy of around 93%, surpassing conventional methods which only achieved 50% correct classification.
Researchers at Princeton University have developed a software tool using machine learning to predict reaction yields, reducing time and cost for synthesizing new medicines. The tool, which can handle up to four reaction components, uses random forest models to accurately forecast yields for thousands of reactions.
Researchers at U.S. Army Research Laboratory and University of Texas at Austin developed a new algorithm called Deep TAMER to train robots using human feedback. The algorithm enables robots to learn tasks in a short amount of time with accurate critique, improving performance on complex tasks like Atari Bowling.
Researchers used machine learning to decode brain activity during a simple task of distinguishing between BA and DA syllables. The results show that the brain uses specific regions for mental associations related to the task, not just for processing information.
Researchers trained machine learning models on radiologist reports to identify clinical concepts in CT scans, achieving an accuracy of 91%. This technology will help develop artificial intelligence to diagnose diseases and improve patient care.
Researchers trained neural networks on thousands of images from simulated high-energy particle collisions to identify key features. The networks achieved up to a 95% success rate in this analysis. Machine learning algorithms will next be applied to actual experimental data to further advance our understanding of the universe's mysteries.
Researchers from Innsbruck and Vienna teams used artificial intelligence to design new quantum experiments, leveraging a projective simulation model and reinforcement learning. The AI-agent performed tens of thousands of experiments, discovering novel structures that could be tested in the lab.
A new algorithm developed by Kirk Bansak et al. uses machine learning and optimal matching to align refugees with suitable employment locations based on their individual skillsets. The results show that the algorithm boosts employment success in both the US and Switzerland, with gains of up to 71% compared to current practices.
Researchers have developed a machine learning model to predict geothermal heat flux beneath the Greenland Ice Sheet, revealing an anomalously high heat flux in northern regions. The study uses 22 geologic variables to improve ice-mass loss and global sea-level rise predictions.
A new method of securely communicating between multiple quantum devices has been developed, enabling a large-scale, un-hackable quantum network. The approach uses quantum laws to ensure security and can work for any device, regardless of manufacturer, bridging the gap between theory and practical implementation.
Researchers developed codes MENNDL and RAVENNA to efficiently design and train neural networks, generating and training up to 18,600 networks simultaneously. This enables the training of highly accurate networks in a fraction of the time, with applications in self-driving cars, intelligent robots, and scientific experiments.
Researchers at Osaka University developed a technique called lexical acquisition through implicit confirmation to enable AI to learn words in the flow of dialogue. This method allows computers to acquire knowledge about unknown words during conversations with humans, making them smarter and more responsive.
Researchers at University of Helsinki develop a new privacy-aware machine learning method that enables accurate modeling using private user device data. This method ensures limited information on each data subject is revealed, making it ideal for protecting sensitive health and human behavior data.