AI-woordenlijst
Het complete woordenboek van kunstmatige intelligentie
Reinforcement Learning for Materials Synthesis
Application of reinforcement learning where AI learns optimal synthesis strategies by interacting with a virtual or real laboratory environment, maximizing the yield and desired properties of new materials.
AI Phase Space Mapping
Use of machine learning algorithms to predict and visualize regions of thermodynamic stability of crystalline phases in multi-component systems, thereby accelerating the discovery of alloys and ceramics.
Automated Gradient Descent on Energy Potential
Optimization technique where AI iteratively adjusts the structural parameters of a material to minimize its total energy, calculated by interatomic potential models, to predict its stable crystal structure.
Knowledge Graphs for Material Properties
Interconnected data structures representing entities (materials, elements, experiments) and their relationships, enabling AI to reason about cause-effect links between composition, structure, and properties to guide discovery.
AI-based Interatomic Potential Model
Computational model, often a neural network or support vector machine, trained on quantum chemistry data to predict the potential energy of an atomic system, offering a trade-off between accuracy and computational speed.
Multi-objective Bayesian Optimization
Efficient exploration strategy of the materials design space that uses a probabilistic model to sample the most promising compositions, balancing exploration of new regions and exploitation of known areas for multiple target properties simultaneously.
Machine Learning Prediction of Thermodynamic Stability
Use of AI models to estimate the formation energy or chemical potential of a material, allowing quick determination of whether it is thermodynamically stable or will decompose into more stable phases.
Graph Neural Network for Crystals
Neural network architecture specifically designed to process crystal structures by representing atoms as nodes and bonds as edges of a graph, efficiently capturing geometric and chemical relationships to predict properties.
AI-guided high-throughput virtual screening
Process of evaluating millions of candidate material compositions using fast AI predictive models, enabling the identification of a small subset of promising compounds for further experimental validation.
Cross-domain transfer learning for materials
Methodology where a pre-trained AI model on a large dataset of one material class (e.g., oxides) is reused and fine-tuned to predict the properties of another less-documented class (e.g., nitrides), reducing the need for data.
Self-supervised learning for crystal structures
Training technique where the model learns relevant representations of crystal structures from automatically generated pretext tasks (e.g., predicting masked atoms in a unit cell), without requiring expensive property labels.
Generation of novel structures by VAE
Use of a Variational Autoencoder (VAE) trained on known crystal structures to learn their latent distribution, and then to sample this space and generate new geometrically valid and potentially stable crystal structures.
Language Model for Chemical Formulas
Application of Transformer-type models (like BERT) to scientific text corpora to learn contextual relationships between chemical compounds, their properties, and their synthesis processes, in order to infer new knowledge.
Band gap prediction using convolutional neural networks
Use of convolutional neural networks (CNNs) applied to 2D or 3D representations of a material's electronic structure (like charge density) to directly predict its band gap, a key property for electronic and optical applications.
Co-optimization of composition and microstructure
AI approach that simultaneously treats chemical composition and microstructure parameters (grain size, texture, etc.) as input variables to discover materials with optimal performance, recognizing their interdependence.
Uncertainty analysis for property predictions
Integration of techniques that quantify model confidence (e.g., Bayesian networks, model ensembles) into material property predictions, allowing to prioritize experiments on the most promising and most uncertain candidates.