AI & 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.
매일 00·06·12·18시(한국시간)에 웹에서 새 소식을 찾습니다. AI 활용 방법과 결과가 확인된 새 사례만 게시합니다.
Checking notifications…
An ACS Nano paper applied machine learning regression and SHAP analysis to automated perovskite thin-film fabrication data to identify environmental parameters that strongly affect thin-film quality and proposed data-cleaning and feedback strategies. This is a case that can be directly connected to real-time data interpretation and process control in materials processing.
It was reported that the transformer-based diffusion model IDPForge generates all-atom conformational ensembles of intrinsically disordered proteins (IDP) and intrinsically disordered regions (IDR), and produces ensembles that agree well with experimental solution data. It was also explained that sampling with experimental restraints incorporated into the process worked with no additional computational cost.
The paper published in JACS on 2026-10-30 is identified as a case that used hierarchical machine learning to interpret site heterogeneity in nanoparticle catalysts. For materials researchers, this can be read as an approach that decomposes different active sites and differences in reactivity on catalyst surfaces in a data-driven way.
A method has been reported for applying a physics-informed neural network to low-dose atomic electron tomography to more robustly and completely reconstruct 3D atomic structures. AI for materials analysis was in fact used to address the difficulty of structure determination under low-dose conditions.
OpenAI announced on 2026-10-27 that it is expanding its collaboration with Oracle to connect Oracle applications and data to ChatGPT. OpenAI models and products will also power Oracle's new products and services, targeting enterprise workflows.
A study has been reported in which generative AI was directly used to design single-layer diverse chiral metasurfaces. It is useful for inverse design and candidate generation, but manufacturability and performance validation must follow.
An approach has been reported that uses a large language model to automatically extract standardized experimental preparation procedures from literature. It can be used to convert materials and chemistry literature into structured procedure data, but general-purpose performance and error rate remain challenges.
Anthropic said it has granted the independent evaluation organization METR a broader level of access, so it can more strongly verify model safety and risk assessments. This is an extension of the company’s earlier commitment to strengthen external evaluation, and it can be read as part of a trend toward improving the effectiveness of third-party validation.
Google DeepMind has unveiled Genie 4 and begun providing it first to trusted testers in the form of a research preview. As a world model that emphasizes real-time interactive simulation and multimodal generation capabilities, it is notable in that AI is becoming more concrete in the way it creates and explores virtual environments.
The paper published in ACS Applied Materials & Interfaces on 2026-10-28 is an example of using an AI-assisted simulation framework to interpret the structure and spectroscopic signals of Pt/γ-Al2O3. The specific algorithm and quantitative performance are not included in the currently confirmed information.
The paper published in J. Phys. Chem. Lett. on 2026-10-28 is a study that predicts the bandgap of hybrid halide perovskite monolayers with machine learning. As far as currently available information shows, the specific model type and quantitative performance are not disclosed, but it can be regarded as an example of applying ML to materials property prediction.
OpenAI announced GPT-6.1 Terra on 2026-10-26 and said it would be available in the API, ChatGPT Work, and Codex as gpt-6.1-terra. The model supports vision reasoning, computer use, and writing tasks, and is not included in general ChatGPT chat.
The JACS paper rapidly evaluated 28 co-solvent candidates for aqueous zinc batteries using machine learning molecular dynamics(MLMD), reporting about a 10^4-fold speedup compared with ab initio molecular dynamics(AIMD). In this screening, N,N-dimethylacetamide(DMAC) was proposed as the optimal candidate. The key point is not battery operation itself, but that co-solvent discovery was accelerated with AI.
A paper accepted by Physical Review Materials demonstrated more efficient learning by estimating candidate-structure uncertainty for noncollinear spin-lattice potentials with equivariant evidential deep-learning-based active learning. The uncertainty score was strongly correlated with prediction error, and the select-train-fine-tune procedure reduced energy, force, and projected spin-force errors compared with random sampling.
Lila Sciences reportedly evaluated 2,942 catalyst compositions using a closed-loop laboratory and machine learning models, and identified 6 families of palladium-based oxides that could reduce dependence on iridium in electrolytic hydrogen production. However, actual durability in industrial electrolyzers remains a separate challenge.