Deep Learning

A novel deep learning architecture for multi-source data fusion

"Reading the invisible": POSTECH-led team develops AI framework accounting for hidden defects in metal 3D printing

Noninvasive brain-computer interfaces are bringing robotic assistance closer to everyday life

Penn researchers create AI tool to speed antibiotic discovery

Deep learning helps engineer stronger continuous fiber composites

Researchers developed a deep learning method that optimizes both material layout and fiber direction for continuous fiber composite structures, reducing design time by up to 99.7% while maintaining strong accuracy. The ResUNet-GAN framework produces high-performance structural layouts directly from design parameters.

Tiny gestures, big signals: AI learns to read hidden emotion

Micro-gesture recognition is emerging as a key area of research in affective computing, focused on analyzing subtle, involuntary body movements that may reflect hidden emotional states. The field has expanded from RGB video to skeleton, audio, text, and privacy-preserving multimodal settings.

Deep learning extends global nighttime light history

Researchers have developed a deep-learning framework to reconstruct a global, high-resolution nighttime light dataset from 1992 to 2024. The new product improves upon existing datasets by reducing saturation-related bias and better capturing temporal changes in urbanization, economic activity, and human development.

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.

NIH-funded AI model predicts cancer survival from single-cell tumor data

Researchers developed a cancer assessment tool that can identify high-risk patients and specific cell populations linked to their risk. The tool, called scSurvival, predicts survival outcomes more accurately than traditional methods by analyzing single-cell data at cellular resolution.

New transfer-learning model could improve real-world EV charging duration prediction

Researchers propose a novel SENet-CNN-Transformer model to predict electric vehicle charging duration, outperforming existing models in accuracy and reducing training time. The approach combines data enhancement, channel attention, convolutional neural networks, Transformer modeling, and transfer learning to address real-world data sca...

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.

New reinforcement learning strategy could make electric bus V2G services more economical

Researchers developed a health-aware V2G strategy using reinforcement learning to optimize charging and discharging times, resulting in significant lifecycle cost savings ($1,539) and extended battery life (21 months). The study suggests electric bus charging stations can be promising platforms for scalable V2G services.

New learning-based motion planning policy could make intelligent vehicles drive more personally

Researchers propose a personalized longitudinal motion planning policy combining reinforcement learning and imitation learning for intelligent vehicles. The approach adapts driving style to target drivers while meeting performance requirements, promoting human-like behavior and increasing acceptance.

New deep reinforcement learning framework could improve eco-driving for hybrid electric vehicles

Researchers propose an integrated eco-driving framework using deep reinforcement learning to optimize motion trajectory planning and energy management. The framework achieves substantial improvements in transverse-longitudinal comfort, energy economy, and power system health, while reducing hydrogen consumption and driving costs.

New AI approach could improve railway fastener defect detection for smarter maintenance

Researchers evaluate the effectiveness of Vision Transformers and convolutional neural networks for faster and more accurate defect detection in railway track fasteners. The study finds that transformer-based models outperform traditional CNNs, suggesting their potential value for predictive health management in rail networks.

Study reveals hidden damage in stony corals using 3D imaging and AI

Researchers used 3D imaging and artificial intelligence to analyze the microscopic structure of coral skeletons, revealing subtle changes caused by Stony Coral Tissue Loss Disease. The study found that Attention U-Net performed best in detecting differences between healthy and diseased corals.

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%.

Penn researchers use AI to surface unreported GLP-1 side effects in Reddit posts

Researchers identified patient-reported symptoms associated with GLP-1s, including menstrual changes, fatigue, and temperature-related complaints, that may not be fully captured in clinical trials or drug labeling. Nearly 4% of Reddit users reported reproductive symptoms, and fatigue was the second most common complaint.

New AI technology to speed drug development

Scientists at the University of Virginia Health System have developed a suite of AI-powered tools, called YuelDesign, YuelPocket and YuelBond, to transform how new drugs are created. These tools can design drug molecules tailored to fit their protein targets exactly, even accounting for protein flexibility.

Scientists develop spatiotemporal correlation-based deep learning framework for bias correction of atmospheric and oceanic variables

How transformers are rewiring graph-based recommendation

Researchers surveyed how transformers are integrated into graph-based recommender systems, improving handling of long-range patterns, sparse interactions, and complex heterogeneous data. The study proposes a taxonomy and design strategies for transformer-based recommender systems across various tasks.

How drones can find their way without seeing

Researchers have developed a new artificial intelligence framework called CLAK that enables drones to localize themselves in GPS-denied environments using non-visual sensors such as LiDAR, barometric altitude, and inertial measurements. The model improves localization accuracy while remaining lightweight enough for practical deployment.

Eureka! Scientists develop new way to detect breakthroughs in science

A team of researchers at Binghamton University has developed a method to pinpoint discoveries that reshaped the course of science. The new metric uses neural embedding to analyze approximately 55 million scientific papers and patents, identifying major breakthroughs and simultaneous discoveries with greater accuracy.

KTU researchers develop a model that improves machine understanding of the real world

A new model combines multiple ways of analysing 3D data, integrating local and global perspectives to interpret complex environments more reliably. The system improves detection of small or partially visible objects in real-world situations, enhancing safety in autonomous systems.

Can AI learn to read ancient pottery the way an archaeologist does?

A new deep learning model classifies Japanese Sue ware from 3D scans with high accuracy, using three-dimensional point clouds directly. The model achieved an overall accuracy of 93.2%, performing almost perfectly on visually distinct categories, while focusing on regions that may correspond to expert archaeologists' considerations.

Ancient alphabets, new insights: Researchers uncover hidden links among the letters

Researchers from SDSU discovered surprising similarities among ancient writing systems from Africa and the Caucasus region. The study suggests the Armenian alphabet may be more closely related to the ancient Ethiopic writing system than previously thought, revealing possible cultural contact and influence between regions.

Geneva becomes world’s capital of AI in July for ITU’s AI for Good Global Summit

The International Telecommunication Union (ITU) will host the seventh AI for Good Global Summit from 7 to 10 July 2026 at Geneva’s Palexpo convention centre. The summit aims to guide the future of artificial intelligence and unlock its potential to serve humanity.

Researchers pioneer new technique to stop LLMs from giving users unsafe responses

Researchers at North Carolina State University have identified key components in large language models that ensure safe responses to user queries. They've developed a new technique to improve LLM safety while minimizing the alignment tax, which allows AI systems to provide safe responses without affecting performance.

Study: New system aims to detect percentage of recycled plastic in plastic products

Researchers created a new method combining scientific tests and artificial intelligence to differentiate recycled plastic from new plastic. The tool, developed by University at Buffalo researchers, can analyze samples and predict the percentage of recycled content with over 97% accuracy.

MSU study demonstrates faster discovery of therapeutic drugs through AI

A team of researchers at MSU used machine learning to predict how chemicals will influence gene expression, leading to the discovery of promising compounds for the treatment of liver cancer and a chronic lung disease. The study results from years of interdisciplinary work across multiple disciplines and institutes.

Using AI to improve standard-of-care cardiac imaging

Researchers developed a new multiview DNN structure to capture complex 3D anatomy and physiology from multiple imaging views, improving diagnostic accuracy for cardiovascular conditions. The approach demonstrated better performance than single-view DNNs and provided a viable alternative for other medical imaging modalities.

Computational model measures key aging metric from routine biopsies

Researchers developed a computational tool that infers telomere length from structural changes in cells and tissues captured in medical biopsies. The TLPath model accurately predicts telomere length, providing new opportunities for studying human aging.

New deep learning framework solves the cold-start problem

A new framework, DUPGT-CDR, uses gating networks to effectively incorporate both positive and negative feedback in cross-domain recommendation systems, achieving lower prediction errors and improved convergence speed. The framework offers more precise product recommendations and personalized learning resources across various domains.

A comprehensive review charts how psychiatry could finally diagnose what it actually treats

Emerging research across conceptual frameworks, biomarker science, digital phenotyping, and artificial intelligence synthesizes a translational pathway toward a more biologically grounded and clinically useful approach to psychiatric diagnosis. The current system falls short due to standardized clinical language and lack of biological ...

New AI agent could transform how scientists study weather and climate

Researchers developed Zephyrus, an AI agent capable of analyzing and answering questions in natural language about weather and climate data. The agent can handle language-based queries, translating them into code and generating plain language answers.

Ateneo machine learning lab opens doors to industry partners, collaborators

The Ateneo Laboratory for Intelligent Visual Environments (ALIVE) is developing machine learning solutions with industry partners to improve public health, traffic systems, and more. By bridging the gap between messy reality and mathematical models, ALIVE is creating intelligent visual systems that can handle real-world conditions.

Deep learning-enabled virtual multiplexed immunostaining of label-free tissue for vascular invasion assessment

Researchers created a novel approach for simultaneous ERG, PanCK, and H&E image generation from label-free tissue sections, enhancing vascular invasion assessment accuracy and efficiency. The virtual multiplexed immunostaining method overcomes traditional IHC limitations, such as section-to-section variability and tissue loss.

AI accurately spots medical disorder from privacy-conscious hand images

Researchers at Kobe University developed an AI model that can diagnose acromegaly with high sensitivity and specificity using only pictures of the back of the hand and clenched fist. This approach holds promise for disease screening, particularly in rural or resource-constrained areas where access to specialists may be limited.

Don’t Panic: ‘Humanity’s Last Exam’ has begun

A global consortium created an exam with 2,500 questions spanning multiple subjects to assess AI capabilities. Current AI models consistently fail the exam, highlighting gaps in their understanding. The project aims to provide a long-term benchmark for evaluating advanced AI systems and demonstrate the importance of human expertise

Bar-Ilan University and NVIDIA researchers improve AI’s ability to understand spatial instructions

Researchers developed a new method called Learn-to-Steer, which analyzes internal attention patterns of image-generation models to guide their placement according to user instructions. The approach improved accuracy in understanding spatial relationships by up to 61% in existing trained models.

DEGU debuts with better AI predictions and explanations

Researchers have developed DEGU, a tool that improves the accuracy and efficiency of deep neural networks in predicting genomic experiment results. DEGU reduces the size of models while maintaining predictive capabilities, making it easier to understand uncertainty and drive reliable discoveries.

With the right prompts, AI chatbots analyze big data accurately

Researchers at UCSF and Wayne State University found that generative AI tools can perform orders of magnitude faster than human teams in analyzing health data. Junior researchers paired with AI generated viable prediction models in minutes, outperforming experienced programmers in hours or days.

SeaCast revolutionizes Mediterranean Sea forecasting with AI-powered speed and accuracy

SeaCast consistently outperforms the Copernicus operational model over a 10-day forecast horizon and extends predictions to 15 days, generating forecasts in just 20 seconds using a single GPU. This advancement enables rapid 'what-if' scenario testing and probabilistic ensemble forecasts.

Enhanced pulmonary nodule detection and classification using artificial intelligence on LIDC-IDRI data

This study developed and evaluated an automatic method for lung nodule detection and classification using a CNN-based architecture on the LIDC-IDRI database. The proposed method achieved high sensitivity and accuracy, with competitive performance compared to recent studies.

Turning down the heat

A University of Houston professor has found that tree-like thin films release heat at least three times better than traditional methods, enabling more efficient cooling in AI data centers. The discovery demonstrates the power of physics-aware AI design for validating high-impact cooling solutions.

AI and brain control: A new system identifies animal behavior and instantly shuts down the neurons responsible

Researchers at Nagoya University developed an AI system called YORU that recognizes animal behaviors with over 90% accuracy. The system combines real-time video capture with optogenetics to selectively target brain cells driving specific behaviors, offering a major breakthrough in social behavior studies.

A smarter way for AI to understand text and images

Researchers at UC San Diego developed a new training method for AI systems to improve their performance in solving complex problems that require both text and image interpretation. The approach evaluates the quality of training data and grades models based on their logical reasoning, reducing the risk of incorrect interpretations.

New AI model improves accuracy of food contamination detection

Researchers at Oregon State University have developed a deep learning-based model for rapid bacterial contamination detection, eliminating misclassifications of food debris. The enhanced model can reliably detect bacteria in three hours and has the potential to prevent outbreaks and protect consumer health.

Powering AI from space, at scale

Researchers have developed a passive, solar-powered orbital data center that can scale AI computing and reduce environmental impact. The system leverages decades of research on 'tethers' and could host thousands of computing nodes to replicate terrestrial data centers.

Should companies replace human workers with robots? New study takes a closer look

A recent study from Binghamton University School of Management reveals that focusing on human-robot collaboration can generate additional economic value and improve a company's ability to capture a greater share of the competitive market. By leveraging robots in collaborative settings, organizations can foster a positive sense of commi...

Chungnam National University develops AI model to accelerate defect-based material design

Researchers at Chungnam National University have developed an AI model that uses deep learning to predict stable defect configurations in materials. The model, trained on data generated by conventional simulations, can generate results in milliseconds rather than hours, accelerating the material design process.

The Institute of Science and Technology Austria (ISTA) receives €5 million donation for AI research

The Institute of Science and Technology Austria (ISTA) has received a significant donation to advance trustworthy AI technology. The €5 million gift from Garrett Camp will support fundamental research in artificial intelligence, focusing on interdisciplinary collaboration and long-term impact.

Stress-testing AI vision systems: Rethinking how adversarial images are generated

Researchers created a new frequency-aware approach to crafting adversarial images that better match human visual perception. The method, called Input-Frequency Adaptive Adversarial Perturbation (IFAP), significantly outperformed existing techniques in structural and textural similarity.

Creative talent: has AI knocked humans out?

A large-scale study reveals that generative AI models have reached the threshold of average human creativity, but the most creative individuals still outperform even the best AI systems. The study also highlights the importance of human guidance and parameterization in modulating AI creativity.

AI4Min-PE: Predicting reactivity of deep-Earth substances

A new AI model, ShapKAN, has been integrated into the cloud-based platform AI4Min-PE to predict five critical thermodynamic parameters across extreme pressures. The platform offers speed and accuracy far beyond conventional approaches, enabling scientists to explore the chemistry of materials under extreme conditions.

Physics of foam strangely resembles AI training

Engineers at the University of Pennsylvania have discovered that foams exhibit internal motion resembling deep learning in AI systems. The study suggests a common mathematical principle underlying both foams and AI training, with implications for designing adaptive materials and understanding biological structures.

Deep learning model trained with stage II colorectal cancer whole slide images identifies features associated with risk of recurrence – with higher success rate than clinical prognostic parameters

A deep learning model trained on stage II colorectal cancer whole slide images accurately identified features linked to recurrence risk. The study found the model surpassed clinical prognostic parameters in predicting patient outcomes.

Using the physics of radio waves to empower smarter edge devices

Researchers at Duke University have created a new method to use analog radio waves to boost energy-efficient edge AI, enabling devices to run powerful AI models without heavy chips or distant servers. The approach, called Wireless Smart Edge networks (WISE), achieves nearly 96% image classification accuracy while consuming significantl...

AI river forecasts may be accurate—but based on flawed logic

A new study found that popular AI tools for predicting river flow often misinterpret how heat and evaporation affect water, raising concerns for flood and drought planning. The researchers developed a hydrology-specific 'explainable AI' framework to uncover these issues.

AI-powered ECG analysis offers promising path for early detection of chronic obstructive pulmonary disease, says Mount Sinai researchers

Researchers at Mount Sinai have developed an AI-powered ECG analysis tool that shows promise in detecting Chronic Obstructive Pulmonary Disease (COPD) early. The model achieved high accuracy rates across diverse populations, including a subgroup with irregular heartbeat and smoking exposure.