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 from Lehigh University and Seattle University found that making patients aware of biases in human healthcare decisions increases receptiveness to AI recommendations. By highlighting the limitations of human judgment, healthcare providers can create a more balanced relationship between patients and emerging technologies.
A Caltech-led team has developed a control strategy called FALCON that uses reinforcement learning to adaptively learn how turbulent wind can change over time, allowing UAVs to predict and respond to extreme turbulence in real-time. The strategy has been tested in a challenging test setup and shows promising results.
A new Multi-task Learning (MTL) model detects 85% of abusive posts originating from right-leaning individuals on social media platforms.
Researchers have found that integrating machine learning with statistical methods improves disease risk prediction model accuracy. The study highlights the potential of such integrated models in clinical diagnosis and screening practices, which could lead to better patient outcomes.
Researchers at Indiana University are developing next-generation ophthalmoscopes to spot early warning signs of diseases like Alzheimer's, diabetes, and heart disease with a simple eye scan. The technology uses machine learning and AI to reduce diagnosis time from days to minutes.
Researchers from Charité have shown that deep brain stimulation using electrical impulses can accelerate movement and shorten delays in Parkinson's patients. By decoding the intent preceding voluntary movement seconds before action, they discovered that dopamine significantly speeds up this process.
Researchers developed a machine learning model to predict mesenteric lymph node metastasis preoperatively in colorectal cancer patients. The XGB-based model achieved high accuracy, identifying key predictors such as perineural invasion and hematocrit levels.
Researchers have developed new AI models for plasma heating that can predict plasma behavior more accurately than existing numerical codes. The models use machine learning to analyze data generated by a computer code, enabling faster simulations without compromising accuracy.
Researchers have developed AI-enabled detection software that can accurately detect natural debris, litter, or waste blocking culverts. The system can be integrated to existing CCTV systems to provide proactive flood defense, improving safety for response teams.
Researchers developed an electronic tongue that can identify differences in liquids and detect food safety concerns. The AI-powered system achieved high accuracy when using its own assessment parameters, providing insights into the neural network's decision-making process.
Researchers tested 24 MLLMs on Raven's Progressive Matrices, finding that open-source models struggled significantly. However, closed-source models like GPT-4V performed relatively well, suggesting a need for more advanced resources and training data to improve AI's cognitive abilities.
Researchers used LSTM networks to detect cyber threats in SWaT plant industrial control systems, capturing complex time-dependent patterns missed by traditional methods. The study demonstrates the effectiveness of LSTM technology in safeguarding industrial control systems from cyberattacks.
The two-year study aims to explore biases in AI systems and develop a 'human-in-the-loop' framework for quality data discovery. It will investigate how humans can be involved as labelers, prompters, and validators to improve data sets and user interfaces.
A new study of bubbles on electrode surfaces could help improve the efficiency of electrochemical processes by understanding how blocking effects work. The findings show that only a smaller area of direct contact is blocked from its electrochemical activity, not the entire surface shadowed by each bubble.
A Kennesaw State University researcher aims to develop open-source, hands-on QML training materials to educate future researchers. The project will create nine training modules with hands-on labs covering key quantum computing concepts.
Researchers at the University of Tokyo introduce a new optical computing scheme called diffraction casting, which improves upon existing methods. The system uses light waves to perform logic operations and has shown promise in running complex calculations, including those used in machine learning.
The Endocrine Society's inaugural Artificial Intelligence in Healthcare Virtual Summit will explore AI's potential to improve medical care, advance research, and leverage big data. Key sessions will discuss predictive analytics, machine learning algorithms, and natural language processing.
Researchers developed an AI-driven approach to model complex hand movements, overcoming current limitations in neuroscience and biomedical engineering. The model achieved a 100% success rate in controlling virtual Baoding balls, showcasing its strength in various challenging situations.
The new framework, called SLIViT, has been developed by UCLA researchers and achieved accurate disease risk biomarkers detection from medical scans. It consistently achieves better performance compared to domain-specific state-of-the-art models.
Researchers at TU Graz have developed a new machine learning method that generates precise live MRI images of the beating heart using only a few MRI measurement data. This breakthrough enables faster and cheaper MRI applications, including quantitative MRI for diagnoses.
The study challenges the idea of creating artificial general intelligence (AGI) with human-level cognition, citing limitations in replicating human cognition. Researchers argue that even under ideal circumstances, it is impossible to achieve AGI due to the complexity of cognitive processes.
A team of researchers aims to improve autonomous vehicle safety by identifying and mitigating vulnerabilities in software and hardware. They plan to use knowledge gained from a $926,737 NSF award to design protection mechanisms that can be applied selectively to ensure safety while maximizing efficiency.
Researchers at UT Austin developed AI model EvoRank to design protein-based therapies and vaccines by leveraging nature's evolutionary processes. The model identifies useful mutations in proteins, offering a new approach to biomedical research and biotechnology.
A systematic review by PPPL researchers found that most journal articles on machine learning for solving fluid-related PDEs are biased towards machine learning, with negative results underreported. The authors propose rules to make fair comparisons and argue that cultural changes are needed to address systemic problems.
Researchers developed DIAMANTE, a data-centric semantic segmentation approach to detect forest tree dieback events in satellite images. The approach trains a U-Net-like model on labelled remote-sensing datasets and achieves reasonable accuracy for early disease detection, reducing false alarms.
A recent study published in PNAS Nexus found that widespread adoption of large language models like ChatGPT led to a significant decline in user activity on Stack Overflow. The study highlights the impact of ChatGPT on public knowledge sharing and its implications for AI's future.
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 PSU English professor is using a grant from the National Science Foundation to study one of the world's fastest and largest supercomputers, Aurora. The research aims to answer questions about how supercomputers work, their applications in science and industry, and their impact on the US's position in scientific advancement.
Researchers from the University of Xiamen developed a machine learning potential to study Pt-water interfaces, revealing distinct types of water molecules and their anisotropic behavior. This understanding is crucial for elucidating interfacial processes in electrochemical reactions.
Scientists analyzed millions of tweets to detect early warning signs of PTSD in COVID-19 survivors, achieving an accuracy rate of 83.29%. The study highlights the potential for machine learning techniques to identify individuals at risk of PTSD through social media data.
Researchers found that large language models used in home surveillance can make inconsistent decisions about calling the police, even when videos show no crime. Models often disagreed with each other and exhibited inherent biases influenced by neighborhood demographics.
The integration of artificial intelligence in intratumoral immunotherapy can refine diagnoses, guide interventions, predict treatment responses, and adapt therapeutic strategies. This enables the enhancement of patient outcomes, including improved survival rates and quality of life.
Regenstrief researcher Andrew Gonzalez proposes developing an AI chatbot to identify PAD in asymptomatic individuals, providing clinicians with critical information to make informed decisions. The project aims to improve diagnosis and reduce diagnostic errors, particularly among vulnerable populations.
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 aim to advance training methodologies for superintelligent systems, ensuring they learn from imperfect and evolving data. The project seeks to develop new techniques to improve AI reliability, reduce biases, and increase accuracy.
A new knee exoskeleton has been developed to support the quadriceps muscles during lifting tasks, helping workers maintain better posture even when fatigued. The device, which uses a complex algorithm to predict assistance needs, enabled participants to lift faster and with improved posture.
Researchers develop AI-driven catalyst discovery and simulate complex interactions to enhance hydrogen generation, carbon capture, and energy storage efficiency. The project aims to create a knowledgeable and skilled workforce capable of addressing critical challenges in the clean energy transition.
Researchers have found two distinct maps in the brain's secondary motor cortex that enable spatial planning and navigation, with implications for understanding neurological conditions such as stroke. The study discovered a self-centred map used for planning actions and a world-centred map used to determine body position in the world.
A new study published in JAMA Health Forum found that machine learning can be more effective than traditional methods for distributing scarce treatments to patients most vulnerable during a public health crisis. The model reduces expected hospitalizations by about 27 percent compared to actual and observed care.
A new study by Boston University researchers found that the majority of social media posts from e-cigarette brands left out health warnings, despite a federal requirement to include them. The study used AI to analyze over 2,000 Instagram posts and discovered that only 13% complied with FDA health warning requirements.
Researchers have introduced DSFN to improve the speed and accuracy of diagnoses of retinal disorders. This AI-powered medical imaging technique combines retina images with vascular distribution information to accurately locate the fovea in complex clinical scenarios, enabling doctors to detect early signs of ocular diseases.
Researchers from Tokyo Institute of Technology have developed a novel screening methodology using machine learning to identify key design guidelines for ternary metal sulfide electrocatalysts. Focusing on crystal structure leads to better results, overcoming challenges in material properties and electrochemical performance analysis.
The new diagnostic test system combines a field-effect transistor with a paper-based analytical cartridge, achieving over 97% accuracy in measuring cholesterol levels. This innovation has the potential to transform at-home testing and diagnostics with its high sensitivity, low cost, and machine learning capabilities.
Researchers at Klick Labs developed an AI technique using vocal biomarkers to predict chronic high blood pressure with up to 84% accuracy. The study used machine learning to analyze hundreds of indiscernible vocal biomarkers, including pitch variability and speech energy distribution patterns.
A new AI algorithm, DPAD, developed by Maryam Shanechi's lab, can dissociate brain patterns related to specific behaviors, improving brain-computer interfaces for paralyzed patients. The algorithm can also discover new patterns in the brain that may be missed by prior methods.
A deep-learning algorithm developed by astronomer David Harvey can untangle the complex signals of self-interacting dark matter and AGN feedback in galaxy cluster images. The Inception model achieved an accuracy of 80% under ideal conditions, showcasing its potential for analyzing vast amounts of space data.
ISTA's Lisa Bugnet, Alicia Michael, and Marco Mondelli have been awarded ERC Starting Grants to develop new methods for extracting information from data, studying gene regulation, and understanding time-keeping in cells. Their projects aim to simplify data analysis, accelerate personalized medicine, and uncover the secrets of biologica...
A new framework uses multiphysics and machine learning models to predict lithium-ion battery overheating and prevent thermal runaway. This could be integrated into an electric vehicle's battery management system to stop a battery from overheating, protecting drivers and passengers.
Researchers have developed a novel approach using deep learning to accelerate the solution of Navier-Stokes equations, a set of classical equations that describe fluid dynamics. The team's method achieved inference latencies of just 7 milliseconds per input, outperforming traditional finite difference methods.
A Mayo Clinic team developed a computational tool that analyzes the gut microbiome with at least 80% accuracy. The Gut Microbiome Wellness Index 2 identifies subtle changes in gut health, enabling proactive health indicators and potential prevention of chronic diseases.
Researchers at Duke University have developed a computer model that simulates nerve responses to electrical stimulation, enabling the efficient design of more effective and targeted neuromodulation therapies. The new tool, called S-MF, runs thousands of times faster than current industry standards without sacrificing accuracy or detail.
Researchers developed a tool to improve data transparency in large language models, enabling practitioners to find suitable datasets for their models. The tool, Data Provenance Explorer, automatically generates summaries of dataset creators, sources, licenses, and allowable uses.
A team of researchers created RENAISSANCE, an AI-based tool that simplifies the creation of kinetic models to accurately depict metabolic states. The tool successfully generated models that matched experimentally observed metabolic behaviors in Escherichia coli, simulating how the bacteria would adjust their metabolism over time.
Researchers identified abnormal seismic activity three months before two major quakes in Alaska and California. The detection method uses machine learning to analyze datasets derived from earthquake catalogs.
A study by Dr. Shunichi Kasahara found that levels of identification with one's face remain consistent regardless of agency or control over facial movements. The results suggest that a sense of agency does not significantly impact our ability to judge our facial identity, even in scenarios like deepfakes.
The Human AugmentatioN via Dexterity (HAND) center aims to develop robots capable of enhancing human labor through engineered systems of dexterous robotic hands, AI-powered fine motor skills, and human interface. The center's goal is to make robotic assistance accessible and applicable to a wide range of physical actions.
Researchers at the University of Illinois have developed a method to understand and improve light-harvesting molecules for solar energy applications. By combining AI with automated chemical synthesis and experimental validation, they were able to produce molecules four times more stable than traditional ones.
A new study published in Scientific Reports confirms a linear relationship between blood glucose levels and voice fundamental frequency, suggesting potential for voice-based glucose monitoring. Researchers at Klick Labs used vocal biomarkers and AI to detect Type 2 diabetes with high accuracy.
Researchers used AlphaFold2 to predict structural effects of mutations on protein stability, finding correlations between small structural changes and stability changes. This breakthrough opens up new possibilities for protein engineering, enabling scientists to design proteins with specific functions more effectively.