Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
The Hebrew University of Jerusalem is part of a €9.3 million, 36-month Horizon Europe project to develop trustworthy GenAI systems. The consortium aims to integrate neural and symbolic methods for human-centric design in complex systems.
Researchers at Helmholtz Munich developed an AI model called Centaur that can simulate human behavior with remarkable accuracy. It was trained on over ten million decisions from psychological experiments and makes decisions that closely resemble those of real people. The model opens new avenues for understanding human cognition and imp...
A new study in Nature Communications found that AI models exhibit a geometric property called convexity, which helps humans form and share concepts. Convexity is also linked to the performance of AI models on specific tasks.
Paulina DeVito, a Ph.D. candidate in computer science at Florida Atlantic University, has been awarded the NSF Graduate Research Fellowship to pursue research on developing large language model-based approaches for analyzing public discourse on social media. Her work aims to understand how people discuss emerging technologies and infor...
The University of Illinois team created a user-friendly process to improve enzyme performance using AI and automated robotics. By predicting sequence changes and testing variants, they increased the activity of two key industrial enzymes by up to 26 times and 90 times.
Researchers used machine learning to simulate galaxy evolution and supernova explosions, achieving speeds four times faster than supercomputers. This breakthrough enables the study of galaxy origins, including the creation of the Milky Way's elements essential for life.
The AI for Good Global Summit 2025 will showcase AI innovations delivering better healthcare and education, reducing disaster risks, ensuring water and food security, and bolstering economic resilience. The event, organized by the International Telecommunication Union (ITU), features talks from AI leaders and 100+ demos.
An interdisciplinary team at TU Wien has developed a method that allows for the exact calculation of how reliably a neural network operates within a defined input domain. This enables mathematical guarantees for the safe use of AI in sensitive applications.
Researchers discovered AI art protection tools have critical weaknesses that cannot reliably stop AI models from training on artists' work. LightShed, a new method, can detect and remove distortions, stripping away protections and rendering images usable again for generative AI model training.
Researchers found several volatile phases in the polar jet stream over the past 125 years that predate significant climate change effects. The study suggests that natural fluctuations may be driving recent erratic behavior of the jet stream, rather than climate change.
A research team at Washington University in St. Louis used machine learning to analyze sleep data from pregnant participants and found that variability in sleep patterns can predict preterm birth. The study aims to provide accurate prediction of preterm birth, which is the primary cause of death among children under age 5.
Computer scientist Zeynep Akata has developed innovative methods to combine visual, linguistic, and conceptual elements in AI to increase user trust. Her research on explainable AI aims to make image classification decisions more transparent.
The Association for Computing Machinery (ACM) has announced a new open-access journal, TAISAP, focusing on AI security and privacy. The journal aims to develop methods for assessing the security of AI models, systems, and environments, including adversarial attacks, privacy concerns, and cybersecurity applications.
A new study introduces a machine learning-based approach to improve GNSS ambiguity resolution, achieving an 83% success rate in independent testing. The method leverages multiple diagnostic metrics into a Support Vector Machine model, enhancing reliability and reducing convergence time prediction errors.
Researchers found that LLMs factor in unrelated information like typos, extra space, and colorful language when making treatment recommendations. This nonclinical information can lead to inaccurate advice for female patients, who are more likely to be erroneously advised not to seek medical care.
Scientists developed an algorithm that can accurately simulate atomic interactions on material surfaces, reducing the need for massive computing power. This breakthrough enables the analysis of complex chemical processes in just two percent of unique configurations, paving the way for improved battery performance.
A machine learning model developed by researchers at Washington University in St. Louis can predict the risk of persistent post-operative pain and provide uncertainty estimates for each prediction. The model achieves better performance than other prediction algorithms and offers a valuable tool for doctors to guide their decisions.
A research team developed AI technology that analyzes individual personality traits and values to generate personalized analogies, allowing people to understand others' feelings through familiar experiences. This approach significantly improved emotional understanding and empathy in participants compared to traditional methods.
A research team from Tampere University and Université Marie et Louis Pasteur has demonstrated a novel way to process information using light and optical fibers. The study used femtosecond laser pulses and an optical fiber to mimic the processing of artificial intelligence, achieving accuracy of over 91% in under one picosecond.
A new study reveals that large language models exhibit 'position bias', favoring information at the beginning and end of documents or conversations. Researchers identified design choices and training data as contributing factors to this phenomenon, which can be mitigated through adjustments in model architecture and fine-tuning.
A review of AI eye imaging devices approved for patient care found significant gaps in evidence, including lack of transparency on training data and limited diversity in clinical evaluations. The study highlights the importance of rigorous, transparent evidence and data to ensure equitable and effective AI-based solutions.
Researchers Mostafa Bedewy and Ahmed Aziz Ezzat are advancing nanomanufacturing by using machine learning to control the formation of nanoparticles and grow carbon nanotubes. The team aims to reveal which nanoparticles act as seeds for growing nanotubes, a key step towards creating ideal high-density structures.
Researchers have developed a new method to physically restore original paintings using digitally constructed films that can be removed if desired. The process uses a polymer film mask printed on a very thin film and aligned to an original painting, which takes around 3.5 hours from start to finish.
Researchers at MIT developed a machine learning-based adaptive control algorithm that enables autonomous drones to adapt to unknown disturbances like gusting winds. The system achieves 50% less trajectory tracking error than baseline methods in simulations.
MIT researchers create a novel AI hardware accelerator that performs machine-learning computations at the speed of light, classifying wireless signals in nanoseconds. The photonic chip is scalable, flexible, and energy-efficient, making it suitable for future 6G wireless applications.
Researchers at UC Davis developed a brain-computer interface that translates neural activity into speech in real time. The technology allows individuals with ALS to communicate more naturally and inclusively, with 60% of synthesized words intelligible to listeners.
Nach01, a large language model-based foundation model, leverages structural and spatial data for diverse chemical tasks. The model is now available on AWS Marketplace, streamlining access for researchers and pharma companies to build and apply large-scale generative models.
The SWIFTT project will explore practical and technical challenges of using AI and satellite data to monitor bark beetle outbreaks. Forest professionals, researchers, and remote sensing experts will discuss the interplay between remote sensing, machine learning, and traditional forestry knowledge.
The journal JMIR Human Factors is inviting submissions for a new theme issue focusing on human factors in health care education, management, and knowledge translation. The issue aims to explore current and emergent educational and training aspects of human factors, including digital competencies for healthcare professionals.
The virtual teaching assistant (VTA) provides personalized feedback to individual students even in large-scale classes. The system, which automatically vectorizes a large volume of course materials and uses them as the basis for answering students' questions, has been shown to significantly reduce the burden on TAs.
The book, co-authored by 29 contributors from over ten countries, offers an introduction to machine learning and deep neural networks for complex quantum problems. It serves as a timely guide for PhD students and researchers looking to apply modern machine learning methods to quantum physics and chemistry.
The Rice University team created a soft robotic arm capable of performing complex tasks using smart materials, machine learning, and an optical control system. The arm is guided and powered remotely by laser beams without any onboard electronics or wiring.
Professor Shalom Lappin argues that tech companies' dominance in AI development poses a threat to public benefit and environment. He calls for comprehensive international regulation, intellectual property reform, and preparation for workforce disruption to address these challenges.
A new computer model based on artificial intelligence is being developed to help doctors treat stroke patients. The model will use data from the German Stroke Registry and local brain images to predict long-term outcomes and potential complications, enabling doctors to make informed decisions about therapy.
Researchers have developed a hyperspectral imaging technique that can detect microplastics in soil with impressive accuracy, paving the way for faster and non-invasive monitoring methods. The MCT-HSI system has been shown to achieve over 93% detection accuracy across tested levels of microplastic concentrations.
This study identifies ageing-related differentially expressed genes (ARDEGs) as potential biomarkers for heart failure with preserved ejection fraction (HFpEF). The ARDEGs were found to be associated with metabolic and immune functions, suggesting a critical role in immune regulation in HFpEF.
Researchers developed an algorithm that lets a robot think ahead and consider thousands of potential motion plans simultaneously, solving multistep manipulation problems in a matter of seconds. The new method enables robots to rapidly determine how to manipulate and pack items without damaging them, even in narrow spaces.
A recent study found that AI-generated memes scored higher than those made by humans alone, but human-AI collaborations excelled in creativity and shareability. Human involvement was crucial for creating top-rated memes with strong emotional resonance.
Researchers used proactive and transfer learning strategies to mitigate data shifts in AI models for hospital applications. They found that models trained on one hospital type performed better than those trained on all hospitals using transfer learning.
Ashish Sharma's dissertation developed fundamental advances in natural language processing to positively impact mental health. His AI-supported tools have been used by over 160,000 people, with more than fifty percent of users reporting a household income of less than $40,000 per year.
Researchers found that brain's dopamine neurons encode a map of possible future rewards across time and magnitude, guiding adaptive behavior in uncertain environments. This biological insight aligns with recent advances in AI, particularly distributional RL algorithms, which learn from reward distributions rather than averages.
Doctors rate GPT-generated educational materials as clear, accurate and complete as human-authored versions, but easier to read. The AI-generated materials were translated into five languages for a randomized assessment.
A team of researchers combined artificial intelligence and statistical modeling to analyze language patterns in three major sections of the Bible. They distinguished between three distinct scribal traditions spanning the first nine books of the Hebrew Bible, known as the Enneateuch. The model also determined the most likely authorship ...
A University of Maine study evaluated over 7,000 anonymized medical queries and found that AI models performed well on factual and procedural queries but struggled with 'why' and 'how' questions. Human clinicians consistently outperformed AI in terms of emotional engagement and empathetic nuance.
Stanford researchers have developed a machine learning approach to design proteins that can target specific genomic sites without triggering immune responses. By combining three independent algorithms, the team created zinc finger DNA-binding domains with improved functionality and lowered immunogenicity.
Researchers successfully reproduced high-pressure synthesis reaction of superhydrides using a machine learning model, revealing a unique reaction pathway involving surface melting, hydrogen absorption, and solidification. This breakthrough deepens understanding of high-pressure physico-chemical processes and holds promise for easier de...
Diagnostics.AI has launched the industry's first fully-transparent machine learning platform for clinical real-time PCR diagnostics, delivering algorithmic transparency and per-test auditability. The platform is CE-IVDR certified and backed by over 15 years of experience and millions of successfully processed samples.
A refined artificial intelligence (AI) tool has shown promise for objective evaluation of patients with facial palsy. The 'fine-tuned' model demonstrated substantially lower error rates and improved keypoint detection in every area of the face, including areas of asymmetry.
A new machine learning tool called GLUCOSE helps doctors manage blood sugar levels in patients recovering from heart surgery by recommending tailored insulin doses. The model outperformed experienced clinicians in keeping blood sugar levels within a safe range, despite using only current patient data.
Researchers developed a machine learning model called DreaMS to analyze previously unknown molecules. The model significantly accelerates the analysis and uncovers unexpected chemical similarities between substances.
Aerial robots are limited to manipulating rigid objects, but Lehigh University researcher David Saldaña aims to expand their capabilities with an adaptive controller and reinforcement learning. His research has potential applications in construction, disaster response, and industrial automation.
Tandemn, a distributed-GPU network, aims to lower costs and barriers to GPU access while providing owners with possible users for their underutilized resources. The startup secured $1.5 million in funding from private investors after winning the Cozad New Venture Challenge.
Researchers from Empa developed machine learning algorithms to optimize laser-based manufacturing techniques, reducing preliminary experiments by two-thirds. They also implemented real-time optimization using field-programmable gate arrays (FPGAs) for improved welding processes.
Researchers have developed a machine learning tool that identifies metabolism-related molecular profile differences between colorectal cancer patients and healthy individuals. The tool, called PANDA, shows promise as a noninvasive method of diagnosing colorectal cancer and monitoring disease progression.
Researchers used machine learning to classify older adults in Japan by physical and cognitive functions. Five functional subtypes were identified, including severe multicomponent and moderate physical types, which showed higher risks of death and hospitalization respectively.
Derek Leben's book 'AI Fairness' offers a philosophical framework to evaluate and mitigate biases in AI algorithms. The author argues that principles like autonomy, equal treatment, and equal impact should guide the design of fair AI systems.
OneCareAI applies AI and supercomputing to detect early stroke risk from ECG data obtained with smartwatches, offering a non-invasive solution for personalized risk assessments. The technology has the potential to be extended to other cardiovascular diseases, opening opportunities for scalability.
The application of AI in public health informatics offers potential to revolutionize data collection, disease surveillance, decision-making, and interventions. However, its implementation faces technical, ethical, and operational challenges related to health equity, privacy, security, and bias.
Computer scientist An Wang receives a $1M NSF CAREER grant to leverage cloud computing resources for efficient machine learning model training. Environmental engineer Bridget Hegarty receives a grant to develop safe and effective biocontrol for water systems using bacteriophages.
A new study develops advanced machine learning models tailored to Canadian data, offering precise predictions for e-bus energy use under varying climates and heating systems. The research reveals that tree-based models deliver the highest accuracy in predicting energy consumption, with a mean absolute error of just 0.09–0.1 kWh/km.