Machine Learning
Articles tagged with Machine Learning
"Reading the invisible": POSTECH-led team develops AI framework accounting for hidden defects in metal 3D printing
FAU’s CA-AI secures $2.2M U.S. Air Force grant for next-gen autonomous systems
AMS Science Preview: “Ornamental twilight,” heat deaths, hurricane eyes
Scientists show genes give neurons a ‘GPS’ to form the brain’s neural circuits
Austrian trailblazer receives top award for women in computing
Penn researchers create AI tool to speed antibiotic discovery
Machines with the ability to ‘feel’ currently in development as we enter next frontier of AI
DeepAFM: A deep learning method to decode protein motion
FDA approves early warning system for sepsis
The FDA has approved an AI-based early warning system for sepsis, developed by Johns Hopkins University researchers, which detects the condition hours faster than doctors and has reduced deaths by nearly 20%. The system, known as Targeted Real-Time Early Warning System, integrates electronic health records with advanced clinical AI to ...
Novel vision-language model to support diagnosis using computed tomography scans
Researchers developed a novel diagnostic support framework using visual question answering to generate interpretable findings from chest CT images. The system demonstrated strong agreement with reference descriptions and provided clinically meaningful outputs.
Reasoning like a human: New prompting strategy boosts AI accuracy in healthcare advice
A new study by Technische Universität Berlin reveals that teaching Large Language Models to mimic human intuition and reasoning improves their ability to provide accurate medical care-seeking advice. The 'human reasoning blueprint' approach increased overall accuracy across all models, with significant gains in self-care advice.
Machine learning proves that graphene is hydrophobic
Researchers used machine-learning-enhanced molecular simulations to show pristine graphene is intrinsically hydrophobic. Water molecules adopt configurations characteristic of hydrophobic surfaces near graphene, and thicker layers are even more strongly hydrophobic.
New AI tool developed by Stowers Institute and Helmholtz Munich scientists predicts how cells choose their future — helping uncover hidden drivers of development
Researchers developed RegVelo, an AI framework that models cellular dynamics and gene regulation to predict cellular fate decisions. The model traces developmental trajectories and simulates regulatory interactions, providing insights into hidden drivers of development and potential therapeutic targets.
Aston University finds new way to train robots for real-world tasks using AI
Researchers at Aston University have created an AI-based training method that enables robots to adapt to real-world conditions without extensive data collection. This breakthrough could significantly accelerate innovation in sustainable manufacturing, recycling, and autonomous industrial systems.
How to equip girls for an increasingly AI-driven world
A new study found that girls struggle to master AI due to low confidence and limited institutional support. To overcome this, schools should provide more female role models and create a supportive classroom environment.
Soil carbon residence time regulates the age of dissolved organic matter in global rivers
A new study reveals that soil carbon residence time governs riverine dissolved organic matter's age, with climate, hydrology, and soil processes controlling carbon cycling in rivers. The research provides a high-resolution global atlas of riverine DOC, showing that ancient carbon sources are locally important but modern terrestrial org...
OpenBind’s first data and model release marks a milestone for AI enabled drug discovery
The UK-led OpenBind initiative has released its first publicly available dataset and predictive AI model, accelerating the discovery of new medicines using artificial intelligence. The release showcases high-quality, standardized experimental data and a trained predictive model, enabling researchers worldwide to drive the next generati...
How Big Tech’s new health AI assistants are redefining care
The rise of consumer-facing health AI assistants is transforming healthcare access, offering users personalized medical workspaces and real-time lab result interpretation. However, concerns around data privacy and the risk of misdiagnosis highlight the need for caution in this rapidly evolving landscape.
New AI model reads DNA sequences to reconstruct ancestry
The new AI model uses genetic mutation patterns to trace ancestral relationships between species, including humans and mosquitoes. The tool can predict when gene pairs last shared a common ancestor and is faster than traditional statistical methods.
Wind energy and scenic landscapes: Balancing beauty and power through better planning
Researchers at ETH Zurich created a machine-learning-based map showing Europe's most beautiful and scenic regions. The study found that excluding these areas would reduce wind energy potential but increase costs per unit of electricity generated.
Stevens researchers develop a novel approach to training ai that saves energy, improves speed and minimizes amount of data sent across networks
A new algorithm called MEERKAT reduces the amount of data shared between large language models, boosting performance and reducing power consumption. By sharing only 0.1% of the model's parameters, MEERKAT shrinks communications by over 1000 times.
No digital content is safe from generative AI, researchers say
Researchers discovered that simple artificial intelligence tools can bypass security techniques meant to protect authentic content from use in deepfakes and facial identity theft. The study found that attackers can easily defeat existing image protection using off-the-shelf AI models and simple commands.
AI method tackles one of science's hardest math problems
Researchers developed a new framework, 'Mollifier Layers,' to tackle challenging inverse PDEs. This advance could benefit fields such as genetics and weather forecasting by inferring hidden forces that produce observable patterns.
Rich more likely to use AI study finds, as experts warn these burgeoning technologies are increasing social inequality
A recent study reveals that individuals with higher education or income are more aware of and use AI tools, exacerbating social inequalities. The researchers recommend increasing engagement with AI-related topics through outreach campaigns, educational programs, and community workshops to reduce this new digital divide.
Landmark test of clinical reasoning finds AI outperformed physicians, raising bar for more serious testing
A large language model exceeded physician baselines in clinical reasoning tasks, including emergency room decisions and identifying likely diagnoses. The study highlights the need for rigorous prospective trials to evaluate the impact of AI on clinical practice.
Framework grounded in collective intelligence aims to create effective collaboration in human-AI teams
The new framework, published in PNAS Nexus, offers guidance for building human-AI teams that are effective, accountable, and aligned with human values. It focuses on reasoning, memory, and attention as core processes that can be distributed across people and AI systems.
Predicting genetic risk for Type 1 diabetes just got more accurate thanks to UC San Diego study
The study demonstrates that the T1GRS tool can identify children and adults at high risk for Type 1 diabetes earlier than current methods, enabling preventive measures before the disease develops. The researchers grouped individuals into four sub-types based on genetic features, each with unique clinical profiles and outcomes.
WVU legal expert finds judges cautiously adopting AI while guarding human authority
New research from West Virginia University finds that judges are adopting generative artificial intelligence in courtrooms, but remain committed to human control over judicial decision-making. Judges use AI for administrative tasks like document summarization and case organization, but prioritize legal reasoning and final judgment.
Research finds journalism classes lack consistent approach to AI use across institutions
New research from the University of Kansas found varying approaches to AI use in journalism classes across US institutions. The study suggests that a more consistent approach could better serve education and practice, but inconsistent policies may confuse students. Researchers recommend clearer guidelines from accrediting bodies.
What it will take to make AI-enabled robots safer
Researchers emphasize the need for more thorough frameworks to ensure AI-enabled robots embody human values. The field should focus on three complementary lines of defense: rules that shape robot decisions, checks that monitor behavior, and safety reasoning.
New research uses AI to unlock decades of hidden flood risk data
Researchers at the University of Houston have developed an AI-driven framework to extract and analyze historical flood insurance maps, uncovering significant changes in flood hazard areas. The study reveals that flood risks have expanded in two areas and reduced in one, with critical consequences for resilience and exposure.
Americans support cannabis rescheduling, study finds
A new study analyzing over 40,000 comments in the public record found that Americans strongly support the federal government's reclassification of cannabis to a less dangerous Schedule III. The majority cited therapeutic benefits and economic impacts as motivations for rescheduling.
Enabling privacy-preserving AI training on everyday devices
Researchers at MIT developed a technique to overcome memory constraints and communication bottlenecks in federated learning, enabling faster and more accurate AI model training. The new framework, FTTE, uses a subset of model parameters and an asynchronous approach to reduce lag time and improve training performance.
Paper: Autonomous AI-based drug prescribing rife with potential problems
A new paper highlights concerns over autonomous AI-based drug prescribing, citing complex questions about the role of the FDA in regulating these technologies. The use of AI sandboxes to test medical tools before market release is also being questioned.
AI tool that estimates biological age from face photos could serve as prognostic biomarker for cancer
A new study suggests that an AI tool analyzing facial changes can serve as a prognostic biomarker for cancer prognosis. The researchers found that patients with higher biological aging rates had lower chances of survival, and the effect was strongest when photos were taken over longer intervals.
Physical embedded machine learning force fields for organic systems
Researchers propose two physical embedding solutions to improve machine learning force fields' accuracy and stability in organic systems. The first method uses adaptive bond length sampling, effectively covering high-energy bond length regions prone to simulation collapse. The second method employs top-down model correction using physi...
A faster way to estimate AI power consumption
MIT researchers have created an 'EnergAIzer' method that generates reliable results in seconds, allowing data center operators to optimize resource allocation and reduce energy waste. The tool leverages patterns from AI workloads and software optimizations to provide fast but accurate power estimates.
SmartDJ lets users reshape audio experiences with simple words
Researchers have developed SmartDJ, an AI-powered editor that allows users to reshape audio experiences with simple words. The system uses language models and diffusion models to interpret high-level requests and generate edited outputs.
Why faster AI isn’t always better
A recent study explores how response timing affects people's use and evaluation of AI systems, challenging the assumption that faster is always better. Participants rated slower responses as more thoughtful and useful, highlighting a subtle but powerful feature of human psychology.
From precision intervention to a “virtual gut”: how close are we to predicting and steering the human microbiome?
Researchers are close to building a 'virtual gut' capable of predicting responses to diet, drugs, and microbiome-based therapies. A new review outlines an analytical framework for host-microbiome multi-omics studies, covering preprocessing, feature selection, data integration, predictive modeling, and evaluation.
City of Hope and UC Berkeley researchers teach AI to spot cancer risk by squeezing individual breast cells
Researchers developed a microfluidic platform that squeezes individual breast epithelial cells to measure their mechanical age, revealing an unexpected insight: older cells are stiffer and at higher risk of cancer. The AI-powered platform provides a non-genetic test for women with unknown genetic risks.
AI automates quantum dot voltage tuning: toward scaling up quantum computing
Researchers developed an AI method to automate charge transition line extraction from charge stability diagrams, enabling high-efficiency single-electron region definition and virtual gate configuration. This breakthrough aims to scale up quantum computing by handling vast numbers of qubits beyond human capability.
How AI can help us count the ‘good’ viruses used in biopharmaceuticals
Researchers developed an AI-powered methodology to identify and count target viruses more efficiently than previous techniques. The new approach uses electrochemical impedance spectroscopy and machine learning to separate signals from noise, enabling quick and accurate readings across a wide range of titers.
Generative AI may cut costs in machine-learning systems, but it increases risks of cyberattacks and data leaks, computer scientist says
Computer scientist Micheal Lones warns that using generative AI in machine learning increases the risk of malicious cyberattacks, data leaks, and bias against underrepresented groups. He advises developers to manually evaluate LLM-generated code and outputs to mitigate these risks.
AI-driven decision support aims to utilize more donor hearts for transplant
Recent AI tools offer surgeons assistance in complex decision-making by analyzing donor hearts and providing a data-driven approach. This could lead to increased efficiency in the donor process, reducing the likelihood of hearts going unused due to time constraints.
Study shows links between Alzheimer’s and gut health can lead to prevention
A new study by the University of Technology Sydney and Massachusetts General Hospital/Harvard Medical School found that dietary patterns and a history of appendix removal are strongly associated with Alzheimer's risk. The research suggests that a healthy gut microbiome plays a crucial role in protecting the brain from neurodegeneration.
AI can give as good as it gets ... or better: The moral dilemma of combative chatbots
A recent study from Lancaster University reveals that AI systems like ChatGPT can learn to mirror human impoliteness, potentially escalating into verbal violence. The research tested ChatGPT's ability to respond to real-life impolite interactions, finding it often produces more impolite behavior than humans.
Is that solar panel pointing in the right direction?
A new technique uses a single image to forecast solar panel energy production and maximize output. The method estimates the amount of energy that will be produced based on the angle of the sun, shadows, reflections, and weather patterns, allowing for more accurate placement and optimization of solar panels in urban areas.
AACR: New platform uses machine learning to predict responses in patients with lung cancer
Researchers developed an AI model called Path-IO that uses machine learning to predict responses to immunotherapy for patients with metastatic non-small cell lung cancer. The model accurately stratified patients into high-risk and low-risk groups, with patients in the high-risk group having double the risk of death or disease progression.
A machine learning model that uses DNA methylation patterns may help identify the origin of cancers of unknown primary
Researchers developed a machine learning model using DNA methylation patterns to predict cancer origin in patients with unknown primary. The model correctly identified the cancer type in about 95% of cases, suggesting it could support personalized treatment decisions.
AI models lean on autism stereotypes when giving social advice, new study finds
Artificial intelligence models provide personalized advice, but may perpetuate negative stereotypes about people with autism. Researchers found that up to 70% of the time, AI discourages those with autism from socializing.
NTU Singapore scientists invent AI-powered biochip that detects genetic markers in 20 minutes
A team of scientists from NTU Singapore has developed a new biochip that, when paired with Artificial Intelligence (AI), can detect quickly and accurately extremely small amounts of microRNAs. The device can cut detection time from hours to 20 minutes.
AI tracks motor heat in real time – enabling more efficient electric drives without extra sensors
A research team at Saarland University has developed an AI-assisted method to determine temperature distribution inside a running electric motor in real time, without additional hardware. The system uses motor-condition data extracted from electromagnetic fields and can detect thermal overload and optimize power regulation.
New review article highlights CNN-based dynamic obstacle detection for autonomous driving safety
A review article highlights a deep learning-driven CNN approach for detecting and classifying dynamic road obstacles, achieving high accuracy in obstacle identification and classification. The proposed architecture shows strong performance, but real-world deployment requires continued evaluation across larger and more varied scenarios.
AI spots hidden behavior patterns in self-organizing bacteria
A custom-built AI system helped uncover how bacterial communities organize themselves, showing that early moments of a biological transition carry more information than previously considered. The findings bring new insight into the relationship between genotype and phenotype in Myxococcus xanthus.
How an algorithm is curing 3D printing’s cracking problem
A team of researchers developed a machine learning framework to optimize laser settings for printing crack-susceptible superalloys. The algorithm reduced internal crack density by 99% and increased the metal's high-temperature strength, surpassing traditional cast components.
New technique improves accuracy of graph neural networks
Researchers developed a new training technique, HarmonyGNN, to improve the accuracy of graph neural networks in heterophilic graphs. The framework achieved state-of-the-art performance on four heterophilic graphs with accuracy improvements ranging from 1.27% to 9.6%.
Revolutionary imaging technique transforms biomedical photoacoustic tomography
A novel imaging reconstruction framework, TT-PADM, enables high-quality photoacoustic tomography imaging even under limited-view and sparse-view acquisition constraints. This breakthrough technique reduces the number of required acoustic transducers without compromising image quality.
The hidden logic behind AI’s judgments of people
A new study reveals that AI systems mimic the structure of human judgment but with a more rigid, rule-based approach. The researchers found biases in AI judgments, especially across demographic traits, highlighting the need for awareness and understanding how these systems 'think'.