PSP-HDC: Graph-Structured Hyperdimensional Computing for Microstructure Prediction
Researchers propose PSP-HDC, a graph-structured hyperdimensional computing framework for predicting process-structure-property (PSP) relationships in multiphoton photoreduction fabrication. The method addresses data sparsity and heterogeneity by encoding a directed PSP graph as an internal prior. A trainable scalar-to-hypervector encoder handles diverse parameter scales and noise. The framework aims to improve data efficiency and explainability over conventional feature-vector models and mechanistic pipelines.
Key facts
- Multiphoton photoreduction enables high-fidelity 3D microstructure fabrication.
- PSP prediction is difficult due to sparse, heterogeneous, interaction-dominated data.
- Conventional feature-vector models suffer from spurious correlations and poor regime transfer.
- Mechanistic pipelines require calibrated submodels unavailable in early development.
- PSP-HDC uses a graph-structured hyperdimensional computing framework.
- A trainable scalar-to-hypervector encoder learns parameter-specific embeddings.
- Sample representations are composed through graph structure.
- The framework is designed for data-efficient and explainable PSP prediction.
Entities
—