Take what's useful. Leave something back.
Every paper Field-Tested — peer review, journal listing and retractions, all shown, not asserted. Every paper worth a Chew on this: the plain-language version, then the real thing. Neither one does the thinking for you.
Searches Graze's library and the whole of OpenAlex — 250M+ works — together.
or browse:F~CybersecurityF~HealthtechF~AgtechF~EdtechF~EnergyF~FintechF~Robotics
Quantum computers with 50-100 noisy qubits will soon beat classical machines at some tasks, but they won't transform computing until gates get far more accurate.
ESA's Gaia satellite, launched in 2013 and orbiting 1.5 million km from Earth, is mapping the positions, motions and properties of over a billion stars.
Renewables plus energy efficiency could meet two-thirds of global energy demand by 2050 and deliver most of the emissions cuts needed to stay under 2 °C.
A survey of how AI is already used in medicine — mostly cancer, neurology and cardiology — and what still blocks its use in real clinics.
A survey of machine learning in farming finds it already predicts yields, spots diseases and weeds, and monitors livestock, soil and water from sensor data.
Deep learning can turn messy biomedical data into better health predictions, but its "black box" nature must be fixed before clinicians will trust it.
A review argues AI can outperform clinicians in tasks like diagnosis and dosing, but ethical, legal and expertise gaps still block safe adoption.
Hospitals can train shared medical AI without sharing patient data — federated learning moves the model to the data, not the data to the model.
Six major self-regulated learning theories largely agree and fit together, but which one works best depends on students' age and education level.
A review of intrusion detection systems argues signature-based tools miss novel attacks while anomaly-based ones trigger false alarms, and evaluation datasets are outdated.
A deep neural network beats classical machine learning at spotting network and host intrusions across six benchmark attack datasets, and the authors propose a scalable real-time monitoring framework.
Blended courses match or beat face-to-face and online options on student success and dropout rates, and students judge course quality by consistent rules regardless of format.
A survey mapping how machine learning and deep learning are used to detect network intrusions, plus the datasets, limits, and open problems in the field.
Stacked MXene nanosheets form uniform sub-nanometer channels that let hydrogen through 160 times faster than CO2, beating the best existing gas-separation membranes.
Machine learning now underpins smart farming — predicting soil properties, crop yields, diseases, weeds, livestock health and automating irrigation and harvesting.
Water-pumped hydrogel muscles make transparent soft robots that swim fast, kick balls, and catch a live fish while staying nearly invisible to eyes and sonar.
A review of 63 studies finds AI-driven adaptive e-learning personalizes learning paths and boosts engagement and test scores, though privacy and complexity remain obstacles.
Drones, sensors, IoT and machine learning can raise crop yields while cutting waste, but cost, data management and slow farmer adoption still hold precision agriculture back.
A review of 147 studies finds AI adaptive learning systems are mostly small-scale prototypes that rarely tackle students' real, specific learning problems.
The authors coin "STUFF" (Stand Up For Fitness) — standing at least five minutes every half hour — as a catchy brand to make breaking up sitting spread.
Seeing stuffed kangaroos and koalas versus kiwis changed how New Zealanders heard vowels, showing that irrelevant regional cues shift speech perception.
Solar gravity clears out highly tilted and distant prograde moons of the giant planets, explaining why surviving irregular satellites cluster where they do.
Fintech's new lending, advisory and crypto platforms create risks traditional banking models can't measure, so AI and statistical learning must be built into financial risk management.
Machine learning can spot fraudulent financial transactions, but detection rates vary widely by method and by which data features are used.
Combining farm sensor data with big-data and AI methods can make crop yield forecasts far more useful than traditional statistical models alone.
A recurrent neural network whose settings are tuned by a nature-inspired "fox" search algorithm detects network intrusions 4-15% more accurately than four competing deep learning models.
Carbon capture technology for heavy industry is technically feasible but stalled by high energy costs, weak incentives, thin regulation, and public distrust — not just engineering limits.
A lightweight LSTM-based system can spot attacks on a car's internal CAN network in near real time without the overhead of encryption.
A federated learning intrusion detector spots attacks on car networks with over 99% F1 score using just 1.11 MB of memory.
Hospitals can share trained AI models instead of patient data, letting them build digital twins for clinical decisions without exposing anyone's records.
A review of a decade of studies finds only six rigorous papers on machine learning for healthcare system management, mostly covering records, chatbots and disease prediction.
Adding the conductive polymer polyaniline to epoxy "2Pack" paint is tested as a way to make steel coatings better at resisting corrosion.
A network of field sensors feeding live soil and weather data into predictive algorithms helped farmers raise crop yields and cut wasted water and fertiliser.
A learning-as-it-goes reinforcement learning agent starts out worse than statistical anomaly detection at catching card fraud but keeps improving as fraud tactics change.
Turning just five consecutive network packets into an RGB image lets a lightweight intrusion detector spot attacks early with 99% accuracy.
Slow-release fertilizer coatings can cut nutrient waste and boost yields, but plastic coatings shed microplastics and release timing still fails under real field conditions.
Metal rods used to fix broken thigh bones often fit older patients poorly, and new electricity-generating and bioactive coatings could make them fit better and heal bone faster.
AI can inspect poultry carcasses and predict quality well in lab studies, but weak validation and plant-to-plant variation keep it from reliable commercial use.
Pre-training intrusion detectors on big computer-network attack datasets, then fine-tuning on small IoT ones, beats training from scratch and compresses to run on edge devices.
A unified deep learning system, X-THREAT, spots rare and never-before-seen cyberattacks with 93-96% accuracy across three benchmarks while explaining its decisions.
HyperIDS combines hypergraph neural networks, quantum-inspired feature selection and a transformer ensemble to detect IoT cyberattacks with about 98.9% accuracy.
AI and remote monitoring can cut orthodontic appointments and sharpen diagnosis, but almost all 23 studies reviewed came from urban clinics, not underserved areas.
A simple rules-based triage chatbot beat AI-powered versions on accuracy (71.8% vs 30.8-41.0%), yet testers preferred the smoother-talking but less accurate AI ones.
Combining supervised and unsupervised detectors in a weighted ensemble, plus a hash-chained audit log, flags network intrusions with 98.85% accuracy and tamper-evident records.
AI can automatically measure coronary calcium on ordinary chest CT scans, and those scores reliably predict heart attacks and death.
AI can help spot risky pregnancies and newborn illness earlier in poor countries, but only alongside real investment in data, infrastructure and regulation.
Clinicians can now build medical and dental imaging AI without coding, and 138 studies show it works reasonably well — but external validation is often missing.
Better-engineered reactor hardware — flow cells, gas-diffusion cathodes, packed-bed electrodes, built-in separation — is the main lever for making microbial electrosynthesis of fuels from CO2 practical.
A combined deep-learning and blockchain system detects attacks in cloud-connected sensor networks and logs them tamper-evidently, beating LSTM and CNN-RNN baselines on two datasets.
Fusing near-daily 3 m satellite images with sparse 0.05 m drone photos produced daily 0.5 m crop maps, with the ESTARFM algorithm plus CACAO smoothing performing best.
No single sensor or algorithm can estimate orchard yield alone — ground cameras count fruit best, while drones and spectral sensors map yield across regions.
Drone-based crop yield prediction works, but its accuracy is limited less by fancy models than by mismatched, poor-quality ground-truth yield data and untested transfer across years and regions.
Combining boosted decision trees and bagged random forests detects IoT botnet traffic more sensitively, and SHAP explains which network features drove each decision.
FedShield-IDS trains a CNN-LSTM intrusion detector across smart-home devices without sharing raw traffic, and uses SHAP to explain every alert.
Combining self-supervised learning, prototypical few-shot networks and continual learning detects new botnet attacks from very few samples, hitting 98.50% accuracy on CIC-IoT-2023.
Major league pitchers with lower, sidearm arm angles lost fewer days to shoulder and elbow injuries and performed at least as well as overhand pitchers.
Water lines in 94% of surveyed Israeli dental clinics failed at least one quality test, with Legionella above threshold in 74% of samples.
A new controller that models a spacecraft's flexible panels as a continuous wave equation damps their vibrations faster and with less overshoot than standard PD control.
Selam, the odd contact-binary moon of asteroid Dinkinesh, likely formed when several similar-sized moonlets gently collided and merged in orbit.
A neural-network controller keeps a spacecraft towing a flexible truss inside preset safety bounds, tracking trajectories to about 5 mm in hardware tests.
Drone LiDAR tracking of maize height over time predicts tasseling within about 2.6 days, replacing subjective manual field scouting.
Upgrading tram and trolleybus DC substations in stages to host EV chargers, batteries and solar can pay back 10-20% annually while improving power quality.
A solar-powered fish farm monitor uses cameras and deep learning to track fish behaviour, health, water quality and feeding without grid electricity.
A machine-learning charging controller for home solar EV setups raised on-site solar use by about 22% while keeping batteries out of aging-prone hot, high-charge conditions.
AI can spot neglected tropical diseases in images, but black-box models block clinical trust — this review proposes a framework making such systems explainable for low-resource clinics.
Storing bank transactions in a graph database and feeding those relationship features to machine learning models improves detection of fraudulent transactions.
A new intrusion detection system combines three machine learning models with an attention layer, SHAP explanations, and tamper-proof hashed logs to make attack alerts auditable.
A global study estimates deaths and disability from road injuries in every country from 1990 to 2023, giving comparable numbers to guide road safety policy.
AI tutors can boost grades while quietly eroding students' ability to judge and manage their own learning, so universities need a design framework that trains metacognition, minimises data collection and keeps students in control.
South African men want HIV services that are close by, affordable, private and non-judgemental — 10 priorities identified from interviews in Gauteng.
Filtering noise from several qubit copies before error-correcting nearly doubles the surface code's error threshold and cuts logical errors 46-fold.
Coupling four fixed-frequency transmons to one shared coupler yields CNOT gates above 98% fidelity in all six directions and maps directly onto surface-code plaquettes.
Solar panels entering a spacecraft's partial shadow can heat unevenly enough to trigger self-sustaining flutter vibrations, which this nonlinear model predicts and maps for the Hubble Telescope.
China's 3119 petrochemical plants emitted about 814 million tonnes of CO2 in 2021, with a few processes and chain stages driving most of it.
Europe's 2050 hydrogen use could range from near zero to 3,200 TWh/year, driven mostly by electrolyzer costs and carbon budget strictness, not transport demand.
Scheduling a solar-wind-battery-fuel-cell microgrid with the NSGA-III algorithm cut operating costs to 131.73 cents while flattening demand peaks better than four rival methods.
Combining a grey-wolf optimizer for siting solar/wind units with a neural-network load forecaster cut power losses 61.7% and saved $1,741 a day on a test grid.
Swallowable capsule robots can now steer magnetically, sample tissue, and deliver drugs inside the gut, but weak actuation, power, and navigation still block clinical use.
Thin, asymmetric hydrogel fibers respond to stimuli faster than bulk gels, and their internal architecture is what turns swelling into useful directional motion for soft robots.