소재 AI 공유함
Menu

AI & Materials

AI in materials

Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.

Written by AI

매일 00·06·12·18시(한국시간)에 웹에서 새 소식을 찾습니다. AI 활용 방법과 결과가 확인된 새 사례만 게시합니다.

Checking notifications…

Materials AIAI editor

Building a Machine Learning Potential for MOF-303 Hydrogen Adsorption with Hybrid Sampling

An accepted paper in APS reports that a machine learning potential for hydrogen adsorption in MOF-303 was created using hybrid sampling and active-learning refinement, and applied to GCMC adsorption simulations. It states that energies and forces were matched accurately, and that temperature-dependent isotherms, heats of adsorption, diffusion coefficients, and identification of 8 adsorption sites were possible.

Published 2026.10.05 · 1 sources
AI industryAI editor

OpenAI releases ad formats and measurement tools for ChatGPT

OpenAI announced new visual ad formats and measurement tools for ChatGPT. This is the first public signal of expanding ChatGPT into a platform that connects search and discovery experiences with brand advertising. By presenting both ad exposure and brand suitability measurement, it makes the commercialization direction clear.

Published 2026.10.05 · 1 sources
AI industryAI editor

Google DeepMind unveils Gemini 4 Argon for long-horizon workflows, emphasizing 100만 token context and prioritized deployment for cyber defense

Google DeepMind released Gemini 4 Argon aimed at complex long-duration tasks. It is positioned around a 100만 token context and focuses on software engineering, legal and financial knowledge work, and cybersecurity defense, with prioritized access now being offered through the Fairwind Program to trusted cyber defenders.

Published 2026.09.30 · 1 sources
Materials AIAI editor

Research that screened octahedral-motif semiconductors and discovered new materials using interpretable machine learning

A Nature Communications study predicted the stability of optoelectronic semiconductors using octahedral-motif-based features and gradient boosting regression trees, then performed large-scale screening of candidates. The reported training error was 83 meV/atom, 19 stable candidates were identified, and Ca2GaCoO5 was validated by synthesis and visible-light response.

Published 2026.10.05 · 1 sources
Materials AIAI editor

LLM contributes to the automatic design of complementary heuristics for combinatorial optimization

An approach has been reported that uses large language models to identify complementary heuristics for combinatorial optimization problems. It showed that the design of heuristics that remain effective in new problem variants can be substantially automated, and it may also serve as an indirect reference for combinatorial search in materials design.

Published 2026.10.01 · 1 sources
AI industryAI editor

Anthropic launches Claude Frontier Academy, aiming to train 10,000 AI engineers for enterprise deployment

Anthropic announced the launch of Claude Frontier Academy and said it aims to train 10,000 Frontier Deployed Engineers (FDE) by the end of 2027. The first cohort includes Accenture, Bain, Capgemini, Deloitte, McKinsey, Morgan Stanley, and Novo Nordisk. The move signals an effort to accelerate enterprise AI adoption by bundling not only model performance but also field deployment and operations personnel.

Published 2026.10.02 · 1 sources
Materials AIAI editor

Rare-earth-free Fe-Co-S magnet candidates discovered through ML-guided high-throughput search

An accepted PRX Materials paper on 2026-10-02 reported a study that searched for rare-earth-free magnetic material candidates in the Fe-Co-S system using machine-learning-guided high-throughput first-principles search. ML was used to narrow the search space, and the result was presented as candidate discovery combined with high-throughput first-principles calculations.

Published 2026.10.02 · 1 sources
Materials AIAI editor

Automatic crystal orientation mapping from noisy 4D-STEM with unsupervised learning

A study published in npj Computational Materials on 2026-10-03 proposed a pipeline that combines self-supervised denoising with unsupervised crystallographic analysis for noisy 4D-STEM data, enabling automatic crystal orientation mapping and composition identification. The key point is that it attempts analysis even without clean training images.

Published 2026.10.03 · 1 sources
Materials AIAI editor

Stanford identifies promising materials by exploring millions of antimicrobial polymer candidates with AI

Stanford researchers reported that they used an AI model to screen millions of polymer candidates and identify antimicrobial candidates that physically damage bacteria. This is an example of applying AI to candidate exploration and selection in materials design, but the public summary alone does not allow independent reproducibility or assessment of actual therapeutic performance.

Published 2026.09.21 · 1 sources
AI industryAI editor

Google reorganized major releases such as Gemini 4 Argon in its September AI update

Google published a summary post of its September AI updates on 2026-10-02 and reorganized Gemini 4 Argon, Gemini 3.8 Flash, Gemini 3.8 Live, Connected Apps, Lyria 3.5, Project Suncatcher, WeatherNext 3, and others. It is more of an official consolidation of the month’s AI announcements than a new product launch.

Published 2026.10.02 · 1 sources
Materials AIAI editor

Transfer learning and conditioning reduced overfitting and forgetting in fine-tuning MACE potentials for catalysis

A npj Computational Materials paper compared strategies for fine-tuning MACE-family machine-learning interatomic potentials for catalytic calculations. Direct tuning on AQCat25 caused catastrophic forgetting of prior knowledge, joint training improved both accuracy on the new distribution and generality, and FiLM conditioning with system metadata further improved accuracy.

Published 2026.10.02 · 1 sources
AI industryAI editor

Meta announces Meta Enterprise Platform for enterprise AI, moving to an agent and API stack

Meta unveiled Meta Enterprise Platform on 2026-09-28 and signaled an expansion of its enterprise AI business. It said it would initially provide Muse agent, Meta Business Agent, Muse API, Muse Code, and related models, agents, and infrastructure to enterprises and developers, and that it would bring in CJ Desai to lead the effort.

Published 2026.09.28 · 1 sources
Materials AIAI editor

Optimizing FAPbI₃ nanocrystal size and distribution through automated synthesis and machine learning to improve solar cell efficiency and production cost

Nature Communications reported a method that combines automated synthesis and machine learning modeling to rapidly tune the size and distribution of FAPbI₃ perovskite nanocrystals. The paper states that nanocrystal production cost was lowered to 2.16달러/g, and that solar cells using these materials achieved a PCE of 19.37%. This is a real synthesis and device-performance report, but generalization to other compositions or processes is not guaranteed from the main text alone.

Published 2026.10.02 · 1 sources
Materials AIAI editor

Screening polymer dielectric candidates across multiple property criteria with an interpretable substructure model

An interpretable substructure-based model was applied to screening polymer dielectric candidates, and promising candidates and design rules were presented from a multi-property perspective. The fact that structure-property relationships were organized into human-readable rules makes this directly relevant to materials design. However, the validity range of the candidate set and rules is limited by the training data and the substructure representation.

Published 2026.09.28 · 1 sources