Academic papers, research breakthroughs, and scientific advances in machine learning and AI.
319 articles
Despite the explosive growth of LLMs and their spread into daily life, the underlying Transformer architecture faces fundamental limitations. Researchers are increasingly exploring alternative architectures that could define the next phase of AI development. Source: cacm.acm.org
The Center for Artificial Intelligence and Machine Learning is set to open at the U.S. Coast Guard Academy this fall, aiming to foster research and partnerships in the field. Source: mycg.uscg.mil
Scientists developed a machine learning model to map complex enhancer regions of DNA that regulate gene activity across different tissues. The approach could help researchers better interpret how genetic variations contribute to human disease. Source: news-medical.net
Researchers built an AI-powered system designed to make minimally invasive surgeries safer and faster by quickly aligning X-ray images during procedures. The technology could help surgeons navigate more accurately in real time. Source: news.mit.edu
The new Center for Artificial Intelligence and Machine Learning will be led by CDR Matthew Williams. It is already tied to cutter patrol operations and other Coast Guard activities. Source: executivegov.com
Researchers developed a machine learning approach that identifies cancer cells based on light-scattering patterns in stained cell samples. This could complement traditional cytological screening, where pathologists visually inspect cells under a microscope. Source: medicalxpress.com
Scientists built an AI-assisted computational pipeline combining symmetry-based screening, machine learning, and density functional theory to search for altermagnetic metal-organic frameworks. Out of tens of thousands of candidates, the approach narrowed the pool to just 15 promising materials. Source: azom.com
Stanford researchers introduced Paper2Agent, a system that converts scientific papers into tested MCP servers so AI agents can execute the described methods using natural language commands. The tool aims to let agents reproduce published results and apply them to new datasets. Source: marktechpost.com
Researchers report that applying appropriate logical context to randomized stabilizer tasks achieved perfect accuracy on five- and six-qubit quantum systems. The findings suggest a deeper connection between quantum logic structures and interpretable ML models. Source: quantumzeitgeist.com
As AI systems advance, some researchers are exploring how they can help generate new mathematical problems rather than just solve existing ones. This 'reverse problem generation' approach is presented as a fresh frontier for computational AI in academic mathematics. Source: techxplore.com
Machine learning allows computers to learn patterns from data and use them to make predictions or decisions. In health care, such models are increasingly applied to interpret medical scans, and researchers are now developing new methods to make these AI-driven analyses more dependable. Source: medicalxpress.com
A major review argues that engineers no longer need to build and train costly neural networks just to test an architecture's viability. Traditionally, designing a deep learning model meant an expensive trial-and-error cycle involving hours or days of GPU time. The findings suggest new methods can predict performance before any training happens, potentially saving significant compute resources. Source: bioengineer.org
Researchers from UC Santa Barbara and Lawrence Livermore National Laboratory have launched a joint effort to use AI for accelerating simulations of nuclear fusion plasma. The partnership aims to make complex plasma modeling faster and more efficient, supporting progress toward practical fusion energy. The project was reported by Andrew Masuda. Source: edhat.com
A thin layer of material sitting between the electrolyte and electrodes in lithium-ion batteries plays a major role in how well ions move through the cell. LLNL researchers used machine learning to better understand and improve this interphase layer, aiming to boost battery performance. Source: hpcwire.com
Multi-label classification — assigning several tags to a single object at once — is a common task across modern AI systems. A new approach reframes labels as preferences, making these models more resilient and reliable. Source: bioengineer.org
Industrial control systems silently keep modern infrastructure running — from water treatment to power grids to chemical plants. Researchers are now combining information theory with machine learning to spot cyberattacks targeting these critical systems. Source: bioengineer.org
Proteins drive biological processes, and their function depends on the intricate relationship between amino acid sequence and 3D structure. A new joint model called SimpleDesign aims to design both sequence and structure together rather than treating them as separate steps. Source: machinelearning.apple.com
Google Research introduced ToolGrad, a new framework that generates training data for LLM tool use by working backward from the answer. The approach achieves a 99.8% pass rate and scores 83.1 on the BFCL benchmark. Source: marktechpost.com
Researchers from Columbia University and the University of Cambridge have created a new benchmark to test how well machine learning models predict material properties. The work aims to improve the reliability of AI-driven materials science by combining quantum computing insights with ML evaluation methods. Source: quantumzeitgeist.com
IBM and NASA have released the NASA-IBM Lunar Foundation Model as open source, marking one of the first publicly available foundation models built for lunar science. The model is intended to support research and exploration related to the Moon. Source: newsroom.ibm.com