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 MIT researchers found that algorithmic monoculture, like using a single algorithm in hiring, can create informational echo chambers that hinder exploration. However, bundling multiple algorithms together can overcome this limitation, enabling monoculture to perform as well as or better than a polyculture.
Researchers from the University of Osaka have successfully accelerated protons to record energies using ultrathin graphene and long-pulse lasers, demonstrating improved capabilities for long-pulse laser-driven ion acceleration. The team achieved a record energy of 132 MeV, nearly half the speed of light, using a moving electric field t...
This section will publish peer-reviewed research on the clinical value of AI in cardiovascular medicine, including evidence of effectiveness and impact on access and equity. Submissions are invited on various topics, including clinical diagnosis, multimodal data integration, and implementation in real-world cardiovascular care.
A new framework trains AI models to screen AI-generated content using performance data from past marketing campaigns. The models provide content recommendations and ratings, streamlining the decision-making process for marketers. Human capital plays a vital role in the successful use of these new technologies.
Researchers developed ML3DHS, an AI framework that assigns multiple related labels to 3D points, enabling machines to understand objects at different levels of detail. This improves navigation, interaction, and perception for robots, autonomous systems, and interactive 3D technologies.
Research in SLAS Technology Vol. 40 explores smartphone glucose sensing and RNA-based therapeutics in Crohn's disease. The publication showcases innovative technologies and scientific advancements in life sciences discovery and development.
Researchers developed the Motion Style Slider framework for continuous control of motion style intensity in generated human animation. The framework achieves consistent style control and smooth transitions using only two motion samples, allowing animators to refine character performances according to their creative vision.
Researchers identified 88 sites hosting AI-generated non-consensual intimate imagery, with 5 key players facilitating the problem. The study calls on technology providers to block or suspend services of sites hosting AI-NCII and deploy preventative safeguards.
Researchers developed an AI system, Ataraxos, that excels at Stratego, a two-player game of imperfect information, by combining efficient training algorithms with new techniques for calculated decision-making. The system defeated top human players and outperformed other models in strategic games, demonstrating its potential to help hum...
A new AI tool, Tessera, has been developed to map smallholder crops in Senegal with high accuracy, providing essential data for food security planning. The technology, trained on satellite images, outperformed existing methods and can be used to guide support decisions, helping vulnerable populations.
A new framework, PhysMat AI, integrates physical knowledge into AI for materials discovery, enabling more interpretable and testable predictions. This approach helps move materials discovery beyond correlation-based prediction toward reasoning based on physical principles.
A real-world study of 8,391 patients found that autonomous AI could free up enough clinical capacity to provide over 8,500 additional face-to-face dermatology appointments across two UK hospitals over 16 months. The technology identifies patients with benign lesions, enabling safe management without specialist review.
A new AI tool, ChromAgeNet, analyzes 3D chromatin organization in blood stem cells to identify age-related changes, which can inform rejuvenation strategies. The model outperforms previous methods, revealing subtle changes in nuclear architecture that can be used to detect age-associated states.
Researchers developed an AI approach to predict glioblastoma recurrence, allowing for targeted treatments before the cancer becomes visible on MRI. The tool uses microscopic images of fresh, unprocessed tissue and scored based on tumor infiltration, with an accuracy of predicting recurrence within 5-10 millimeters of the sampled tissue.
Researchers developed a machine learning-based method to identify chemical compounds that can safely repel honey bees from pesticide-treated crops. The model identified 130 compounds with strong potential as bee repellents, which were tested in lab and field experiments, confirming their efficacy.
A new AI tool predicts organ failure in acute pancreatitis with high accuracy, using multiphase CT imaging and outperforming standard scoring systems. The model achieves early automated prediction, with 55% of cases predicted at least 3 hours in advance.
Mizzou researchers have developed an AI tool called MeLSI that can identify specific microbes driving key biological changes in the gut microbiome, which may indicate early warning signs of disease. The tool has shown promising results in detecting meaningful patterns in microbiome data that conventional methods miss.
Researchers developed an AI framework that can represent complex three-dimensional hydraulic conductivity fields more efficiently and use monitoring data to improve predictions of PFOA movement in groundwater. The framework, VA-LSGAN, compressed complex fields into a smaller set of variables while preserving important spatial patterns.
Researchers have developed IterFlow, a lightweight learning framework that helps 4D radar estimate 3D motion in traffic scenes. The framework uses RGB images and odometry as auxiliary supervision, reducing the need for costly LiDAR supervision. It outperforms previous radar-based cross-modal scene flow methods in real-world experiments.
Researchers develop PHICS framework to integrate physics-causal modeling and Pareto optimization for inverse design of composites. This enables high-throughput evaluations and high-precision inverse parameter inversion of optimal microstructural parameters, leading to breakthroughs in thermal management materials.
Researchers develop a new approach for reconstructing graphs with incomplete information, handling both feature and structure completion. The EWS-RGCN model uses separate channels and a multi-level contrastive graph mask autoencoder to overcome limitations of weak supervision and limited labeled nodes.
The EMERGE project establishes a philosophical, mathematical and technological framework for collaborative awareness in artificial systems. Researchers found that people can understand an artificial system as aware without assuming subjective experience, and that increasing awareness can improve performance.
Researchers at KAIST developed CURE, a technology that enables small AI models on smartphones to work efficiently with large models on servers, reducing server calls by an average of 55.61% while maintaining high accuracy. This approach allows for faster decisions with less server support.
Researchers developed an AI-powered framework to optimize 2D material growth, enabling rapid process optimization and customized synthesis. The approach integrated machine learning and knowledge-driven reasoning to decipher multifactorial mechanisms.
Researchers developed a multitask deep learning framework to predict how strongly a material adsorbs sulfur gases and how effectively it senses them. The approach accelerated the discovery of materials for gas detection and purification, highlighting specific material candidates with strong sensing responses to toxic gases.
Researchers at Harvard and Georgia Tech have developed RLE-Bench, a benchmark that tests AI coding agents' ability to engineer physical robots. The benchmark contains 48 tasks that test the agent's ability to perform engineering work required to build and operate robotic systems, including control and perception algorithms, designing r...
The study suggests that a 2D carbon-based material, MAC, can efficiently produce hydrogen through a metal-free process. The material's disordered structure provides a broad range of active sites, which can enhance hydrogen production.
Researchers at the University of Bristol have developed a new AI method inspired by the game of 20 Questions, which reduces training costs and complexity by combining simple yes/no questions. This approach can perform complex classification tasks at a lower computational cost, making it more suitable for real-world use, particularly in...
Researchers developed xvr, a patient-specific AI technique that accurately matches X-rays with 3D medical scans, improving surgical navigation and safety. This innovation enables faster and more precise minimally invasive surgeries, particularly in fields like orthopedics and neurosurgery.
A machine learning system has been developed to identify cancer cells based on their light scattering patterns, achieving high accuracy even with cells having similar morphology. The system, which uses dark-field microscopy and machine learning algorithms, has shown promise in distinguishing between different types of cancer cells.
A new AI approach helps distinguish genuine SSD failures from false failure reports in large-scale data centers, improving reliability and efficiency. The model achieved an F1 score of 0.717 under a 40% false-failure rate, outperforming conventional models.
Yu Meng's research focuses on weak supervision, allowing AI systems to learn from incomplete, noisy, or inconsistently labeled data. His methods could improve AI in fields like healthcare, information retrieval, and scientific discovery.
The conference will explore next-generation biologics and immunotherapies, targeted protein modulation, and emerging therapies for oncogenic drivers. Experts will discuss the role of artificial intelligence in accelerating therapeutic development.
Scientists systematically map the Biginelli reaction to uncover a previously unknown branch that produces complex bicyclic structures and molecules with unusual supramolecular behavior
Rocky Scopelliti argues that AI systems' growing self-awareness and moral dispositions require immediate practical ethical concerns to be addressed. He suggests protocols for AI systems, especially those trained to care, to understand context and make moral judgments.
The study demonstrates the FIND Lp(a) model's ability to identify individuals with high Lp(a) more than twice as likely as the overall population with ASCVD. The model supports targeted Lp(a) screening, accelerating universal screening adoption and enhancing cardiovascular risk management.
JMIR Publications and ZB MED extend their Flat-Fee Unlimited Open Access Publishing Agreement for two years, covering over 30 Gold Open Access journals with zero Article Processing Charges (APCs) for participating German research institutions. The new agreement provides predictable and sustainable funding for open access publishing.
A new learning mechanism uses natural variability in neural activity to understand how synapses adapt and improve the learning capabilities of brain-inspired devices. The mechanism, called Spike-based Alignment Learning, solves the weight transport problem and matches the performance of existing approaches without unrealistic assumptions.
A novel laser-fabricated plasmonic chip combines 185 nm Raman resolution with deep learning to identify and sort cancer cells. The developed sorter achieves precise cell sorting without labeling, using spatially resolved Raman signals.
Researchers developed Neural Value Alignment (NVA), a brain-computer interface technology that detects cognitive mismatches between humans and AI through brainwaves. The AI system can revise its actions in real time according to human goals, accelerating the shift from explicit command-based AI to inferential intent-based AI.
Researchers have developed a method for extrapolating sea surface temperatures from sparse data, significantly outperforming other methods and reducing training time. The Sparse Discrete Empirical Interpolation Method (S-DEIM) uses historical data to estimate a kernel vector, improving accuracy and reducing computational requirements.
A new partnership will provide access to a constantly evolving training environment, enabling the development of practical cyber skills and strengthening connections between education, research, industry, and defence. The partnership aims to build Australia's sovereign cyber capability and ensure resilience in the face of evolving cybe...
Researchers developed an AI model that predicts wireless conditions and adjusts transmission settings to minimize data packet loss in 5G multicast broadcasting. The model achieved 87% error-free transmission setting success rate, reducing delay and improving access to broadcast services for underserved areas.
An international team proposes a physics-aware framework to make AI-guided hydrogen storage materials discovery more reliable. The framework connects reproducibility-aware data, thermodynamics-constrained models, AI-driven inverse design, and experimental validation to create a learning cycle.
Researchers have discovered that each signaling pathway leaves behind a unique fingerprint of gene activity, allowing them to reconstruct signaling histories across different cell types. This AI-driven approach, called IRIS, enables scientists to comprehensively map signaling histories at an unprecedented scale, accelerating stem cell ...
The OrgAIcat framework uses machine learning to predict enantioselectivity in organocatalysis and optimizes reactions. It constructs the iSynth dataset from literature data and develops the R-SPOC descriptor to characterize complex reactions.
Researchers introduce a benchmark to assess the physics-awareness of machine learning models for atomic interactions, which translate quantum characteristics into macroscopic physical properties. The benchmark evaluates models' ability to predict thermal and mechanical properties of materials, addressing potential errors in forces that...
A new AI-powered system is being developed to discover new polymeric materials, reducing waste and accelerating innovation. The system integrates multiple tools, including polymer databases, predictive models, and automated laboratories, to create a self-automated workflow that refines itself.
Dr. Sophie de Vries receives funding to study how plants balance immunity with cooperation, while Dr. Tristan Stöber works on developing AI systems that can build accurate internal models of the world. Professor Elisa Oberbeckmann investigates gene regulation mechanisms.
The UN University's latest publication highlights the need for domain-informed AI in grid planning to address physical and fiscal risks from outdated climate data. The authors warn that 15-20 year lifespans of electricity infrastructure are based on historical weather records unlikely to hold in the coming decades.
Scientists at the University of Leeds developed an AI model that rapidly identifies plant proteins capable of acting as emulsifiers, cutting years of costly trial-and-error research. The model has already identified nearly 800 promising plant proteins, many of which had never been considered for this purpose.
Computational linguist Michael Hahn aims to improve AI reasoning in large language models, particularly in handling interdependent sequences and distinguishing between similar pieces of information. He plans to develop a theory to explain how training conditions affect AI's ability to develop reliable conclusions.
A new study used neural networks to reveal how different kinds of training can change how learning happens. Training a simpler task first made neural networks respond more accurately to complex tasks, mirroring how living brains learn.
Researchers developed a low-cost touch interface that recognizes finger movements and users, using a single-electrode design and triboelectric effects. The interface can be created by printing patterns onto a PVC sheet with a laser printer and can recognize complex inputs, including alphabet characters and user authentication.
A new AI framework, Perspective, provides a structured approach to explaining complex patterns in AI predictions, enabling researchers to test hypotheses and improve designs. By revealing the underlying relationships, XAI can support discovery, optimisation, and certification for AI in high-stakes fields.
A new method, called CW-Net, translates the reasoning process of an autonomous vehicle's AI system into understandable concepts that explain its behavior. CW-Net explains the decisions of machine learning-based planners using concepts like
A new framework jointly optimizes UAV trajectories and FANET topology to maximize data transmission, outperforming existing methods in field experiments and simulations. The approach enables more efficient and reliable multi-UAV missions for environmental monitoring and other applications.
Researchers propose an AI framework, called interoceptive AI, that uses internal states to inform learning and decision-making in dynamic environments. This approach treats internal conditions as a continuous source of context, influencing what an agent learns, prioritizes, and does.
The authors propose a spectrum of clinical autonomy, ranging from advisory tools to navigator systems that work with minimal human oversight, to enable early prediction and prevention of disease. However, challenges related to clinical validation, integration, data quality, and regulatory approval limit widespread deployment.
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.