Learning the Geometry of Materials Space with Trained Neural Tangent Kernels
We use the neural tangent kernel (NTK), evaluated at trained parameters, to recover the geometry learned by a network that predicts superconducting transition temperatures from elemental composition alone.
ProofPaper ↗
Key points
- Neural networks trained to predict materials properties learn representations whose physical content can be difficult to extract.
- Spectral analysis and multidimensional scaling reveal a geometry dominated by a small number of directions.
- Clustering and local neighborhood analysis organize compounds into recognizable superconducting families and connect materials with related chemical and correlated-electron properties despite substantial differences in $Tc$.
- These results establish trained tangent kernels as practical probes of learned materials representations, connecting property prediction with the exploration of materials space.
Sources (1)
- [1]Learning the Geometry of Materials Space with Trained Neural Tangent KernelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 12:07 AM
We use the neural tangent kernel (NTK), evaluated at trained parameters, to recover the geometry learned by a network that predicts superconducting transition temperatures from elemental composition alone.
Neural networks trained to predict materials properties learn representations whose physical content can be difficult to extract.
Extractive summary: sentences quoted from the sources.