zsysuyang commited on
Commit
12c40f5
·
verified ·
1 Parent(s): 2f98e95

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +2 -2
README.md CHANGED
@@ -38,7 +38,7 @@ language:
38
 
39
  ### Challenges
40
  * **Complex Mapping:** The correlation between local atomic structures and spectral features is highly non-linear and high-dimensional.
41
- * **Limitations of Traditional Methods:** Empirical fingerprint matching relies heavily on comprehensive reference libraries, failing for amorphous or complex systems. Meanwhile, *ab initio* calculations (e.g., DFT) provide theoretical rigor but suffer from cubic or exponential computational costs, making them unfeasible for high-throughput screening.
42
 
43
  ### Machine Learning Solutions
44
- To overcome these barriers, **Data-Driven Machine Learning (ML)** has emerged as an efficient alternative for both **forward modeling** (predicting spectra from structure) and **inverse modeling** (inferring structure from spectra).While early ML models relied on manual feature engineering and standard Deep Learning models are often restricted to specific elements (the "element-by-element" strategy), this project aims to bridge the gap. We leverage advanced architectures (such as **Graph Neural Networks**) to capture intra-spectral correlations and strictly structure-spectrum relationships. Our goal is to move beyond single-element constraints and realize a **unified framework** capable of multi-element generalization and multi-target collaborative prediction for large-scale materials discovery.
 
38
 
39
  ### Challenges
40
  * **Complex Mapping:** The correlation between local atomic structures and spectral features is highly non-linear and high-dimensional.
41
+ * **Limitations of Traditional Methods:** Empirical fingerprint matching relies heavily on comprehensive reference libraries, failing for amorphous or complex systems. Meanwhile, *ab initio* calculations provide theoretical rigor but suffer from cubic or exponential computational costs, making them unfeasible for high-throughput screening.
42
 
43
  ### Machine Learning Solutions
44
+ To overcome these barriers, **Data-Driven Machine Learning (ML)** has emerged as an efficient alternative for both **forward modeling** (predicting spectra from structure) and **inverse modeling** (inferring structure from spectra).While early ML models relied on manual feature engineering and standard Deep Learning models are often restricted to specific elements, this project aims to bridge the gap. We leverage advanced architectures to capture intra-spectral correlations and strictly structure-spectrum relationships. Our goal is to move beyond single-element constraints and realize a **unified framework** capable of multi-element generalization and multi-target collaborative prediction for large-scale materials discovery.