Algorithms

Fair matching systems can still produce unequal outcomes, new research finds

A novel deep learning architecture for multi-source data fusion

Researchers “reprogram” materials by quickly rearranging their atoms

‘Nature’s algorithm’ found in Chinese money plants

A team of scientists has found a naturally occurring Voronoi pattern in the Chinese money plant, which helps explain how plants create complex patterns on their leaves. This discovery sheds light on how plants solve problems in nature and may provide new insights into the math underlying evolution and development.

New AI framework Hi4GS revolutionizes wheat yield prediction, boosting accuracy by over 82%

A new AI framework, Hi4GS, revolutionizes wheat yield prediction by improving accuracy and identifying key genetic markers. The framework streamlines high-dimensional genotypic data, uncovering the actual genes influencing yield.

Method for stress-testing cloud computing algorithms helps avoid network failures

Researchers from MIT have developed a more user-friendly and efficient method to identify potential system failures in cloud computing algorithms. The 'MetaEase' technique analyzes an algorithm's source code directly to uncover hidden blind spots that might cause unexpected failures, reducing the risk of costly network outages.

Quantum algorithms for improving surface coatings

Researchers develop quantum algorithms to simulate polymer degradation caused by UV radiation, using industrially relevant aircraft coatings as an example. The goal is to optimize surface coatings for various industries, improving safety and reducing costs.

NUS Medicine partners MitoQ New Zealand to deepen mitochondria-targeted research for healthy ageing

The partnership aims to generate evidence on the potential of MitoQ to slow or improve markers of biological ageing and support longevity. Mitochondria-targeted antioxidants like MitoQ are crucial in producing energy while reducing oxidative stress, a key contributor to ageing.

A new artificial intelligence tool that can generate millions of new molecules

Researchers from Universitat Rovira i Virgili developed an AI tool called CoCoGraph that can generate realistic molecules complying with chemistry laws. The system uses a diffusion model to create plausible structures, resulting in 100% chemically valid molecules, and has been found to be more realistic than other state-of-the-art models.

Medical AI moving faster than safety checks

Flinders University experts caution that AI's impressive capabilities do not automatically translate into safe use for patients. The researchers stress the need for strong governance and clearer standards for evaluation to ensure AI supports doctors in busy care settings.

Researchers create tool to help hunger-relief groups deliver food more efficiently

A new optimization framework helps food banks deliver food more efficiently by accounting for variables such as food availability and household demand. The tool has been incorporated into an app that can also be used by businesses to address delivery logistics challenges.

Transient windstorms pose danger to railroad transport—how can we tackle it?

A recent study has developed an analytical model of downburst wind fields, which reproduces key observable features while adhering to fundamental mechanical principles. The model proposes a framework for assessing train overturning due to downbursts, with high train speeds identified as the most significant contributor to increased risk.

A faster way to estimate AI power consumption

MIT researchers have created an 'EnergAIzer' method that generates reliable results in seconds, allowing data center operators to optimize resource allocation and reduce energy waste. The tool leverages patterns from AI workloads and software optimizations to provide fast but accurate power estimates.

FI-R model, a novel remote sensing method for fine-scale extraction of vegetation

Researchers developed a novel flowering spectral index model, FI-R, for precise fine-scale extraction of angiosperms over large areas. The model achieved high accuracy and applicability across multiple multi-spectral sensor images, with overall accuracies exceeding 94%.

Helping data centers deliver higher performance with less hardware

Researchers developed Sandook, a software-based system that tackles three major sources of performance-hampering variability simultaneously. The two-tier architecture optimizes task distribution for the overall pool while faster schedulers on each SSD react to urgent events.

Evaluating the ethics of autonomous systems

A new testing framework, Scalable Experimental Design for System-level Ethical Testing (SEED-SET), balances measurable outcomes and qualitative values like fairness. The system uses a large language model to capture stakeholder preferences and identifies scenarios where AI systems align with human values.

HEAPGrasp: A faster, smarter way for robots to handle tricky objects using only RGB camera

HeapGrasp uses RGB images to analyze object silhouettes and estimate its 3D shape, reducing the need for depth information. The approach achieves high accuracy while minimizing camera movement and execution time.

Not just faster but smarter: AI that explains its discoveries

A new study developed an AI-driven strategy that accelerates catalyst discovery while revealing the underlying chemistry. The approach, referred to as 'gray-box,' provided meaningful insights into the effect of individual promoters and synergistic interactions between them.

MIT scientists show how the brain handles the “cocktail party problem”

Using a computational model, neuroscientists at MIT showed how the brain selectively focuses attention on one voice among others in a noisy environment. The model found that amplifying the activity of neural processing units that respond to features of a target voice allows that voice to be boosted to the forefront of attention.

Seeing global trade through the lens of physics

Economic complexity methods analyze global networks to generate rankings according to complexity. A new study resolves the uncertainty surrounding these calculations, showing that they lead to a single stable result, with implications for policy-making and analysis of complex networks

A better method for planning complex visual tasks

MIT researchers developed a generative AI-driven approach for planning long-term visual tasks, surpassing existing techniques with a 70% success rate. The system combines vision-language models with formal planners, enabling robots to navigate complex environments and assemble multi-robot teams with high efficiency.

Principles that uniquely determine simple risk-sharing rules

Researchers develop axiomatic approach to understand principles behind simple risk-sharing rules, such as equal sharing and proportional forms. They show how familiar rules can be understood as unique outcomes of particular combinations of axioms, accommodating scenario-based approaches.

A nonparametric framework for inference on integrated quantiles

Researchers developed a nonparametric statistical inference theory for integrals of quantiles, applicable to various data settings. The framework offers unified large-sample arguments for classical estimators and insights into conditions for asymptotic results.

For the Tamarin Prover: Levchin Prize for Real-World Cryptography for CISPA researcher Cas Cremers

Cas Cremers and collaborators receive Levchin Prize for Real-World Cryptography for their work on the Tamarin Prover, an open-source analysis tool for cryptographic protocols, which has had a significant impact on the practice of cryptography and its use in real-world systems.

FAU lands $4.5M US Air Force T-1A Jayhawk flight simulator

Florida Atlantic University has received a $4.5 million grant from the US Air Force to establish a high-fidelity platform for autonomous decision-making and real-time sensor fusion research. The T-1A Jayhawk simulator will be used to study cognitive performance, situational awareness, stress and decision-making under pressure.

Probing entanglement and parameter sensitivity in QAOA via Quantum Fisher Information

A team of researchers uses Quantum Fisher Information to separate individual parameter relevance from cross-parameter correlations in QAOA. This analysis reveals strongly non-uniform parameter relevance that varies with circuit architecture and depth.

New backtests for improved expectile risk forecasts evaluation

Researchers developed new backtesting approaches for expectile forecasts to address issues with existing methods. The new tests disentangle unconditional coverage and independence properties, improving size and power properties.

Predicting extreme rainfall through novel spatial modeling

Researchers developed a new method to predict extreme rainfall in Japan, using Integrated Nested Laplace Approximation - Stochastic Partial Differential Equation (INLA-SPDE), which outperformed traditional kriging methods. The study used hourly precipitation data from 752 meteorological stations across four main islands of Japan and fo...

Digital clinical decision support algorithm substantially reduced antibiotic prescribing without compromising clinical recovery, according to non-randomized controlled trial in 32 Rwandan health centers

A digital clinical decision support algorithm substantially reduced antibiotic prescribing in pediatric outpatient care in Rwanda, according to a non-randomized controlled trial. The study found no compromise in clinical recovery, suggesting the algorithm's effectiveness in guiding appropriate treatment.

Exposing biases, moods, personalities, and abstract concepts hidden in LLMs

A team from MIT and UC San Diego has developed a new method to uncover hidden biases, moods, and abstract concepts in large language models (LLMs). The approach identifies these connections within the model and allows for manipulation of the concept in generated answers.

Big data and human height: ISTA scientists develop algorithm to boost biobank data retrieval & analysis

Researchers from ISTA developed an algorithm that can extract and analyze information from the world’s most extensive biobank with unprecedented accuracy and speed. The method, dubbed gVAMP, enhances the framework's ability to extract complex information from the dataset at hand, providing a detailed overview of the effects on a trait ...

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.

Your social media feed is built to agree with you. What if it didn’t?

A new study from the University of Rochester found that social media algorithms can reinforce echo chambers, but introducing randomness can help reduce this effect. By exposing users to a broader range of perspectives, algorithms can weaken feedback loops and promote more open-minded views.

Big data and LASSO improve health insurance risk prediction

A new study published in Risk Sciences investigates the potential of big data and modern predictor-selection methods to improve health insurance risk assessment. The analysis reveals that adding big data, particularly from smartphone use, improves out-of-sample prediction compared to traditional underwriting information.

Balancing comfort and sustainability with climate-tailored housing

A research team from Osaka Metropolitan University found that optimizing window-to-wall ratio and insulation can reduce energy consumption by up to 27% in subtropical regions. The study provides tailored design guidelines for each climate zone, promoting net-zero energy housing and climate adaptation policies.

Interpretability in deep learning for finance: A case study for the Heston model

A study investigates how interpretable deep learning models can explain neural network predictions for the Heston model. Global methods like Shapley values outperform local methods in producing financially intuitive explanations.

AI and network science for emotional risk measurement in stock markets

A new study uses large language models and network analysis to quantify investor emotions' fragility. High emotional vulnerability is linked to unstable stock prices and lower returns.

Chongqing Medical University team: dual-branch graph attention network enables personalized prediction of ECT efficacy in adolescent depression

A team from Chongqing Medical University developed a model that integrates sMRI and fMRI data to predict treatment responders, achieving high accuracy rates. The dual-branch graph attention network (DBGAN) captures coordinated brain changes linked to emotion regulation and memory processing.

Frontiers in Science Deep Dive series: How breaking the ‘memory wall’ using brain-inspired algorithms could help overcome AI energy costs

Researchers propose a novel approach to AI hardware design by integrating neuromorphic systems and compute-in-memory techniques to overcome the limitations of modern computing hardware. This could lead to more efficient data center energy use and enable real-time intelligence in compact, power-constrained systems.

New research finds Zillow’s Zestimate reduces uncertainty and improves outcomes for both buyers and sellers

New research found that Zestimate boosts efficiency in the residential real estate sector while benefiting lower-income neighborhoods. The algorithm helps alleviate uncertainty about property values, leading to increased buyer surplus and seller profit by an average of 5.94% and 4.36%, respectively.

Integrating credit and debit data for enhanced insights into borrowing behavior and predictive modeling of credit card delinquency

A study published in The Journal of Finance and Data Science shows that integrating credit and debit data enhances the ability to predict credit card delinquency. By analyzing transaction patterns, the model identifies distinct behavioral segments with different risk profiles.

Prognostic tool could help clinicians identify high-risk cancer patients

A new study identifies a practical and powerful prognostic marker that can help clinicians tailor treatment strategies to improve survival for patients with aggressive T-cell lymphoma. Researchers found that patients who relapse within 12 months of initial therapy have a dramatically declining chance of survival, making targeted therap...

A smarter way for large language models to think about hard problems

Researchers developed instance-adaptive scaling framework that uses process reward model to estimate difficulty of question, enabling LLMs to spend more compute on promising solution paths. This approach achieves comparable accuracy with existing methods while reducing computational cost by up to one-half.

Social media research tool can lower political temperature. It could also lead to more user control over algorithms.

A new tool reorders content to downrank posts containing antidemocratic attitudes and partisan animosity, reducing polarization. Users who had such content downranked showed more positive views of the opposing party, with a noticeable impact on their emotional responses.

How personalized algorithms lead to a distorted view of reality

A study suggests personalized algorithms can lead to inaccurate generalizations, even when users know nothing about a topic. Researchers found that algorithmic personalization can bias learning and cause overconfidence in incorrect answers.

How modified robotic prosthetics could help address hip, back problems for amputees

A new algorithm developed by researchers at North Carolina State University can optimize the movement of robotic prosthetic devices and help users exhibit a more natural walking pattern. The algorithm improves hip range of motion and reduces lower back pain for all five study participants.

Enhancing bookkeeper decision support through graph representation learning for bank reconciliation

A new study uses graph representation learning to enhance bookkeeper decision support for bank reconciliation. The method significantly improves match accuracy, especially on one-to-many matches, by leveraging a network of historical records and transactions.

Bigger datasets aren’t always better

The algorithm identifies the minimum set of locations where field studies would guarantee finding the least expensive route, considering problem structure and uncertainty. This method can be applied to broad classes of structured decision-making problems under uncertainty.

More efficient and flexible image compression

Professor Marko Huhtanen's research introduces a new method for compressing images by leveraging the best features of multiple well-known compression methods. The technique enables the removal of rigidity in traditional approaches, allowing for more precise control and adjustment during compression.

Atomic insights could boost chemical manufacturing efficiency

University of Rochester researchers developed algorithms to analyze complex chemistry in propane-to-propylene conversion. The study reveals the importance of defective metal sites and oxide phase stability in catalysts.

A faster problem-solving tool that guarantees feasibility

The new tool, called FSNet, combines machine learning and optimization to find feasible solutions quickly while ensuring constraints are met. It can unravel complex problems several times faster than traditional solvers and even outperform pure machine learning approaches.

New algorithm lets autonomous drones work together to transport heavy, changing payloads

Scientists at TU Delft developed an algorithm allowing multiple autonomous drones to work together to control and transport heavy payloads even in harsh weather. The system enables drones to lift and orient a payload with precision, ideal for reaching infrastructure like offshore wind turbines.

UOsaka breatkthrough: World’s fastest and most accurate self-evolving edge AI for real-time forecasting

Researchers from The University of Osaka developed MicroAdapt, a groundbreaking self-evolving edge AI technology that enables real-time learning and forecasting capabilities within compact devices. It achieves up to 100,000 times faster processing and 60% higher accuracy compared to state-of-the-art deep learning methods.

A special machine for solving NP-complete problems

A team of researchers has developed a new machine called the Electronic Probe Computer (EPC60) that can solve NP-complete problems, including optimal routing, scheduling, and network design. The EPC60 outperforms leading commercial software solvers in solving complex problems with high accuracy and efficiency.

How people learn computer programming

Researchers found that the brain's logical reasoning network was active before learning to code, and continued to engage strongly after acquiring Python skills. This suggests that humans can repurpose cognitive areas involved in reasoning to learn computer programming.

SEOULTECH researchers develop VFF-Net, a revolutionary alternative to backpropagation that transforms AI training

VFF-Net applies label-wise noise labelling, cosine similarity-based contrastive loss, and layer grouping to improve image classification performance compared to conventional forward-forward networks. The algorithm reduces test errors on various datasets, enabling lighter and more brain-like training methods that make AI more sustainable.

Method teaches generative AI models to locate personalized objects

Researchers from MIT and the MIT-IBM Watson AI Lab have introduced a new training method that enables vision-language models to localize personalized objects in a scene. By using carefully prepared video-tracking data with contextual clues, the model is better able to identify the location of a specific object in a new image.

GAN-based solar radiation forecast optimization for satellite communication networks

A novel AI optimization model called GAN-Solar has been developed to address the technical bottleneck of accurate short-term solar forecasting. The model achieves significant improvements on key metrics compared to existing advanced models, producing high-definition forecasts that capture crucial details.

Improved cough-detection tech can help with health monitoring

Researchers developed a more accurate cough-detection model using wearable health monitors' audio and movement data. The new model can distinguish between coughs and nonverbal sounds, improving the accuracy of respiratory disease tracking.

Order from disordered proteins

A team of researchers developed a computational method that can design intrinsically disordered proteins with desired properties. The work uses automatic differentiation to optimize protein sequences and leverages molecular dynamics simulations for precision. This breakthrough has the potential to reveal new insights into diseases like...