Machine learning in computational chemistry

Perspectives on Development of Optoelectronic Materials in Artificial Intelligence Age

“…constructs the two optimal descriptors based on compressive sensing theory SISSO. This method first performs various mathematical operations on the original descriptor to generate a new descriptor, then minimizes the overlap between sample data points to improve classification accuracy [8] . Thirdly, the use of deep learning algorithms such as convolutional neural nets will allow the generation of useful latent descriptors from simple representations of molecule structures.…”

“…This method first performs various mathematical operations on the original descriptor to generate a new descriptor, then minimizes the overlap between sample data points to improve classification accuracy. [8] Thirdly, the use of deep learning algorithms such as convolutional neural nets will allow the generation of useful latent descriptors from simple representations of molecule structures. In order to improve ML model screening ability in perovskite materials for high-efficiency electroluminescent LEDs, more useful molecular information were integrated into the model, such as 208 descriptors from SMILES by RDKit open source, and multiple molecular fingerprints (MACCS, FP2, ECFP).…”

“…This process usually involves design, synthesis and characterization of the material, as well as assembly and optimization of the LEDs. [8] Such a traditional approach requires laborious experimental steps, substantial resource input, and a long research cycle (Figure 1 left). Undoubtedly, a paradigm change in optoelectronics functional materials design and development is in urgent demand for the precise synthesis.…”

“…In fact, the mainstream of LEDs research has been focused on building up the relationship between a new molecular structure and its device performance. This process usually involves design, synthesis and characterization of the material, as well as assembly and optimization of the LEDs [8] . Such a traditional approach requires laborious experimental steps, substantial resource input, and a long research cycle (Figure 1 left).…”