Fig. 2: Optimization performance.
From: Two-step machine learning enables optimized nanoparticle synthesis
a Evolution of the loss for the conditions suggested by RS (green), the BO (blue) and the DNN (orange): each point represents a droplet. For each run, the condition giving the lowest loss is identified as the best performer. The box plots show the distribution of the loss values for all the droplet samples of the run, and the grey zone indicates the minimal loss value to reach before stopping the HTE loop. The absorbance spectra of the BO (b) and DNN (c) best performers, with the associated size distribution of triangular prisms in solution, show that the chosen loss function allows a convergence of the spectra towards the target spectrum and the prisms towards a triangle edge of 65 nm. The colours of the absorbance spectra indicate the colour of the best performing droplet. The TEM images show nanoprisms with the median edge size found in the sample. The scale bars correspond to 50 nm.
