Fig. 4: Schematic illustration showing the clustering and prescreening of each film.
From: Inverse design of chiral functional films by a robotic AI-guided system
a Scheme of the two round clustering-screening process. In the first round, films were hierarchically clustered into several categories as follows: films with high linear dichroism (LD) and transmittance (T0) in which the polarization of the incident light was parallel to the anisotropy were categorized as high performance, and those with both low LD and T0 were categorized as low performance. The probability of each parameter value producing high-performance films (Phigh) and low performance films (Plow) were calculated. If Plow was significantly high (typically >60%), these values were deemed “bad choices” and removed. The second round was similar to the first round, except that the clustering process was replaced by direct classification, in which the low performance ones were defined as either small LD or low T0. b Histogram of LD and T0 for dyed films. The top 40% values are regarded as high performance (highlighted in red shading) and the bottom 40% values are low performance (highlighted in blue shading) in the first round. In the second round, the low thresholds are reduced by 40%. c After hierarchical clustering, dyed films form five clusters, where two are high performance clusters and two are low performance clusters.
