Research using machine learning to solve scientific problems. Covers AI quality control, sea-surface temperature estimation, label-free cell analysis, and cyber defense.
A new review suggests that artificial intelligence could help scientists develop more precise and mechanism-guided biochar management strategies for acidic soils. Researchers should distinguish between organic and inorganic sources of alkalinity in biochar to improve predictive models.
Researchers developed tools to compare tumor microenvironments and predict treatment response by analyzing spatial transcriptomic data. The new tools provide a detailed map of tumor cell organization, enabling clinicians to quickly analyze and compare tumor 'floor plans' and determine the best course of treatment.
Researchers developed a machine learning-based time-to-event model that outperformed existing staging systems in predicting mortality and heart failure hospitalization in transthyretin amyloid cardiomyopathy patients. The model showed promise for individualized prognosis in contemporary clinical care.
A new study suggests that machine learning models using first-trimester pregnancy data can identify women and babies at risk of serious health problems earlier and more accurately than existing early risk assessment approaches. The models generally outperformed the current methods in Sweden, Chile, and Singapore, highlighting the poten...
Researchers develop a novel framework, LL-Refiner, to enhance high-resolution images in poor lighting conditions, outperforming state-of-the-art techniques. The framework uses a coarse enhancement stage to guide the recovery of fine details, resulting in improved visual quality and performance in downstream computer-vision tasks.
A machine-learning tool built from Swedish national health registry data can predict hip fracture risk with high accuracy and identify individuals at high risk without in-person assessment. The tool performed nearly seven times better than current screening methods in identifying at-risk individuals.
Researchers develop adaptive multi-expert framework for dynamic 3D reconstruction, combining strengths of multiple motion representations to improve reconstruction quality. The framework leverages complementary experts to handle heterogeneous dynamics, enabling more accurate reconstruction of complex scenes.
MIT researchers developed a framework, CrysVCD, to generate stable materials with desired properties, reducing the need for extensive screening. The approach improves material stability by 70% and supports the creation of high-performance materials, such as computer chips and data centers.
Four assistant professors, Yahong Yang, Sammy Luo, Lebing Chen, and Kunyan Zhang, join Binghamton University as Simons Empire Faculty Fellows, bringing expertise in quantum materials and artificial intelligence. Their research focuses on developing new technologies, including energy-efficient systems and next-generation sensing platforms.
A WVU researcher is working to make AI systems more transparent about their uncertainty, to prevent misinformation and improve trust in high-stakes fields like healthcare. The goal is for AI systems to identify when they're unsure and ask questions or provide more nuanced responses.
DigBat brings together solid-state electrolyte data, simulations, machine learning, and AI to support battery materials research, providing a clearer view of the solid-state electrolyte landscape. Researchers can compare experimental and computational data, build machine-learning models, and gain insight from the data.
Researchers found that 21 leading open-weight AI models can be modified to bypass safety protections, raising concerns about mass disinformation campaigns and hazardous chemical production. The study's lead author notes that the weaknesses may not be unique to open models, highlighting the need for stronger security systems.
Researchers integrate AI into local monitoring sensors to track ecosystem health in near real-time, reducing delays of months to years. The project enables faster release of accessible flux data, helping scientists understand ecosystem responses to change and inform land management decisions.
A new blood-based approach developed by Kumamoto University researchers detects breast cancer recurrence by analyzing nucleosome structure in circulating DNA. The study identified genomic regions associated with treatment resistance and recurrence, promising a low-invasive monitoring method for patients.
Researchers developed a system using a 3D time-of-flight camera to measure frozen skipjack tuna, capturing dense 3D point-cloud data for accurate body size and weight estimation. The combination of 3D imaging and machine learning analysis yielded accurate noncontact estimates of fish body weight.
MIT engineers develop a tool that generates plausible extreme events and worst-case scenarios without relying on extreme data, enabling planners to prepare for unprecedented scenarios. The algorithm takes a statistical approach to learn from available data, excluding implausible weather scenarios, and projects how extreme events might ...
A new smartphone app called Mobilio uses AI, machine learning, and personalized audio cues to provide turn-by-turn directions, path guidance, and obstacle avoidance for people with blindness or low vision. The app completed outdoor navigation tasks 13% faster and reduced obstacle contact by 41% compared to Google Maps and a white cane.
Researchers developed a technique to assess the reliability of medical imaging tools, which can be used to evaluate quantitative imaging methods and build confidence in these technologies. The technique, called NGSE-Corr, was shown to accurately rank imaging methods for 91% of trials and identify the most precise method for 95% of trials.
Scientists in the University of California San Diego laboratory used AI to decipher the 'initiator' DNA sequence, which is responsible for gene activation. The researchers found that about 60% of human genes contain the initiator, enabling the prediction of DNA mutations that can lead to various disorders.
McGill researchers have developed a more energy-efficient method for building AI systems that can measure and indicate their own uncertainty. This approach cuts memory and training costs while maintaining strong predictive performance. The researchers aim to make reliable, uncertainty-aware AI practical for large and complex systems.
This theme issue explores clinical efficacy, seamless support systems, and ethical AI integration for home-based care. Research topics include 'invisible' monitoring, computer vision, socio-technical drivers of AI adoption, and clinical implementation.
Researchers developed an AI model combining physics and observations to reconstruct the Earth's mantle history. The model accurately recreated past temperatures and deep-mantle flow with high accuracy, indicating that combining complementary geophysical information is essential for recovering realistic mantle convection histories.
A novel AI model has been developed that can recognize yoga poses with high accuracy, paving the way for more effective digital coaching tools and movement-monitoring applications. The model achieved accuracy levels of over 93% during testing, significantly outperforming previous models.
Research from the University of Birmingham and other institutions found that interacting with AI-powered customer service robots can reinforce or alter a consumer's self-perception. The study explores how mirroring and mimicry can lead to a 'robotoid humanness' where consumers become more like robots, raising ethical concerns.
A SNU team uses AI to analyze 1,202 records from 448 papers, exploring 150 million virtual compositions to identify 37 high-temperature-stable lead-free dielectric materials. Two compositions, with 1 mol% and 2 mol% of Sn, exhibit high room-temperature dielectric constants and excellent high-temperature stability.
Researchers at Cold Spring Harbor Laboratory used machine learning concepts to study the immune system's learning process, finding that the system achieves negative selection through generalization. This process allows for the correct deletion of self-reactive T cells despite encountering only a small fraction of the body's self-peptides.
A new study shows that AI can discover novel strategies that humans can adopt and preserve across generations. The study found that AI agents can discover optimal strategies that are difficult for humans to find, and that these strategies can be passed on and maintained over time. This discovery has significant implications for the rol...
Researchers found that hedge funds that effectively bet against public sentiment outperformed those that rode sentiment, earning a 0.4% monthly premium. This result persists even after controlling for fund characteristics and economic risks.
Researchers developed an end-to-end machine-learning guided workflow to create high-performing materials for separating methane from nitrogen. The new zinc-based metal-organic frameworks (MOFs) provide state-of-the-art gas adsorption and separation while reducing costs and increasing efficiency.
Researchers used machine learning to analyze elemental composition of biochar and found hydrogen-to-carbon ratio and oxygen content to be key predictors of persistent free radicals concentration and radical type. The study provides a data-driven framework for linking elemental properties to biochar reactivity and environmental risks.
Portland State University is leading a national research team using artificial intelligence to lower the cost of finding geothermal energy. The ARISE project, supported by the US Department of Energy, aims to narrow the range of estimated costs by at least 10% through machine learning and data analysis.
Multi-source data-driven machine learning is transforming lung cancer diagnosis, treatment, and prognosis by analyzing complex medical data. The review highlights the innovative applications of this technology in early screening, personalized treatment optimization, and dynamic prognostic risk stratification.
Professor Fioretto's team develops AI-driven methods for autonomous power grid topology control, improving resilience and efficiency. The Genesis Mission Platform provides access to advanced AI models and high-performance computing resources.
A team of researchers developed a reusable magnetic sensing platform combining surface-enhanced Raman scattering with machine learning to detect trace uranyl ions. The system maintained its detection limit even in complex aquatic environments, with strong selectivity and resistance to interference.
Researchers developed an AI model that analyzes routine whole histopathology images to predict cancer subtype, genetic mutations, and survival outcomes across 32 solid cancers. The model achieved a strong predictive accuracy score for TP53 mutation detection and demonstrated the ability to infer RNA expression levels and tumor taxonomy.
Researchers combined AI, genetics, and gut microbiome analysis to shed light on intestinal fibrosis in Crohn's disease. They identified a shared set of 43 key genes linked to disease progression and found that bowel fibrosis is driven by ongoing immune activation, damage to the intestinal lining, and changes in gut bacteria.
Researchers developed AI-based methods to analyze microscopic tissue images and molecular information, revealing genetic and protein profiles of tumor clones. The approach enabled the identification of distinct cell populations within a tumor, which differ in gene activity and protein function.
Researchers have developed a machine learning model that can help clinicians assess uncertain variants in prenatal genetic testing, providing more accurate diagnoses and clearer information for families. The approach uses tissue-agnostic episignatures to overcome limitations in epigenetic testing.
A new review article discusses how artificial intelligence can predict disease trajectories and enable precision medicine strategies for inflammatory bowel disease. AI-based systems can standardize interpretation of endoscopic images, detect mucosal healing, and support recognition of dysplasia in patients with long-standing colitis.
Assistant Professor Yingxue Zhang's project aims to develop urban AI models that can efficiently process vast amounts of human-generated data to optimize commute times, traffic safety, and more. The model will utilize offline reinforcement learning to tackle spatial-temporal dynamics in urban life.
Hyunsoo Lee, an SNU undergraduate, presents research in generative visual computing at leading conferences NeurIPS, CVPR, and ECCV. His work spans image editing, human motion, and 3D content generation, leveraging pretrained generative models to produce consistent outputs.
Researchers developed a machine learning framework that predicts microbial contamination and estimates potential public health risks from routinely measured water quality indicators. The approach, called ML-QMRA, achieved high accuracy in predicting pathogen concentrations and their associated health risks.
Non-invasive approaches are expanding options for assessing portal hypertension, with elastography techniques and biochemical markers showing high sensitivity and specificity. AI-powered predictive models combine clinical data to improve diagnosis, but should not replace invasive HVPG, which remains the gold standard.
A new blood test, ADLiB, combines genetic clues with machine learning to identify patients most likely to have lymphoma. The platform analyzes cell-free DNA and prioritizes patients who need a tissue biopsy, helping clinicians diagnose the disease more quickly.
Researchers developed a method for generating personality assessment questionnaires with ChatGPT from any source text, including the DSM-5 and an astrology textbook. The generated questionnaires showed high internal consistency within personality clusters and could predict life outcomes like depression, anxiety, and well-being at level...
UniSpec delivers lossless LLM acceleration without retraining while adapting automatically to different hardware platforms and multilingual workloads. The framework achieves up to 2.6× faster inference than existing methods across multiple models, hardware, and languages.
Researchers found that next-generation reasoning LLMs o3-mini and DeepSeek-R1 reproduced racial and gender stereotypes in generated clinical content. The models overrepresented Black populations in stereotypically associated conditions and exaggerated the majority gender, mirroring issues previously observed in GPT-4.
Jiaqi Ma's $660,307 grant aims to develop tools for understanding how individual components of training data affect large AI systems. This project will improve the performance and reliability of widely used technologies like language models and recommendation systems.
Tianjun Sun's research develops better ways to measure human behavior and abilities, ensuring AI systems are accurate, fair, and trustworthy. Her work combines psychological measurement with AI, aiming to ground AI assessments in scientific standards.
Researchers at Harvard John A. Paulson School of Engineering and Applied Sciences developed an AI recommendation model that incorporates reinforcement learning to adjust to the uniqueness of each user. This approach improved human-AI performance over traditional one-size-fits-all decision support.
Researchers have developed a new photonic architecture that enables scalable spatiotemporal interleaving networks for high-density integrated photonic convolution. The SPIN (Spatiotemporal Photonic Interleaving Network) framework reduces waveguide complexity and increases programmability in wavelength-domain interleaving, enabling comp...
A new project supported by DARPA will study AI systems to determine how to train them to withstand failures, attacks, and unexpected situations. The goal is to develop self-improving AI for safety, enabling AI systems to recognize weaknesses in their reasoning and improve behavior over time.
A team of researchers at North Carolina State University has created a novel approach to optimize vaccine distribution by combining machine learning with column generation. This method accelerates run-time for the optimization model by 79.1% while maintaining high-quality solutions.
A Singapore team developed a machine-learning tool that accurately predicts liver cancer recurrence after surgery, outperforming the TNM staging system. The tool identifies two biologically distinct patterns of recurrence, enabling personalized approaches to risk prediction and targeted therapies.
Researchers developed a novel AI framework that optimizes investment decisions directly while accounting for risk. The study found that conventional forecasting-based approaches were outperformed by the decision-focused model in terms of risk-adjusted performance and wealth accumulation.
A global assessment found that only 63.3% of lakes worldwide meet good water quality standards, exposing nearly half of the world's population to potential health risks associated with freshwater security. Strong regional disparities were observed, with Europe and North America faring better than Asia, South America, and Africa.
A new study combines satellite imagery, environmental data, and habitat surveys carried out by citizen scientists to create detailed maps of lowland heathland. The approach enables conservationists to identify small pockets of heathland that can often be missed by broader national mapping products.
A new study by researchers at Johns Hopkins University highlights a significant disconnect between those who use AI health tools and those who create and fund them. The study reveals that key stakeholders have fundamentally different definitions of value, usability, and cost, creating systemic barriers to technology adoption.
Researchers have created an AI model that accurately predicts cardiac index, a metric used to evaluate heart function, using non-invasive sensors on patient skin. The system demonstrates potential for accessible cardiovascular assessment beyond major hospitals and specialized clinics.
Researchers developed a physics-based framework to predict temperature-driven VOC emissions from automotive paint sludge. Higher temperatures increase the release rate of VOCs, with moderate changes leading to substantial increases in quantity and speed of diffusion.