Articles tagged with Algorithms
A team of researchers from the University of Cambridge and UC Santa Barbara developed 'adversarial' mathematical systems to map out where AI prediction breaks down. They identified two main reasons why machine learning fails: algorithmic limitations and hidden patterns in complex systems.
Researchers developed an auditing technique to test generative AI models for malicious capabilities, without prompting them for illegal outputs. The new approach identifies model variations that can generate CSAM with 100% accuracy.
Using the YEARS diagnostic algorithm is as safe as computed tomographic pulmonary angiography (CTPA) for diagnosing suspected pulmonary embolism in patients with cancer. This strategy saved 22% of patients from undergoing CTPA, highlighting its potential to streamline cancer care.
Researchers developed a new system called Murakkab to optimize agentic workflows for AI applications. It enables developers to describe their intent in high-level terms, automating the selection of models and tools to use, and optimizing hardware configurations in real-time.
MIT researchers developed a framework that allows users to apply constraints to algorithmically generated structures, making them more buildable. The approach has the potential to reduce carbon emissions in construction by up to 90% by designing structures with multiple materials and taking into account materials' properties.
A team of researchers from The University of Osaka has developed a new approach for depth reconstruction from defocus, estimating distances by analyzing blur in an image. Their method combines a coded-aperture camera with diffusion-model-based AI to accurately estimate depth and produce high-quality images.
Researchers have developed a wearable sensor that reads chemical signatures of human breath to decode silent speech into text. The device uses a microscopic nanoforest to capture rapid water vapor changes, achieving 98.51% accuracy rate.
Researchers developed an Insect Synergy Circuit that integrates body movement and internal physiological information to guide insect navigation. The system achieved high accuracy in classifying environmental conditions, enabling gentle control over the insect's movements.
A new DGMoE framework enhances EEG-based emotion recognition by modeling individual differences, achieving high accuracy rates on public datasets. The framework's two-stage selection mechanism and graph-convolution-based expert modules improve robustness and generalization to unseen subjects.
A new open-source trajectory-planning system, MIGHTY, has been developed by researchers at MIT and the University of Pennsylvania. The system enables robots to generate smooth flight paths while reacting to obstacles in real-time, making it suitable for applications such as search-and-rescue, last-mile delivery, and industrial inspection.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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.
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%.
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 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.
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.
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.
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.
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
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.
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.
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.
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.
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.
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.
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...
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.
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.
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 ...
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.
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.
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.
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.
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.
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.
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.
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 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.
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.
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...
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.
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.
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.
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.
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.
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.
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.
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.