Carbon dots (CDs) have wide application potentials in optoelectronic devices, biology, medicine, chemical sensors, and quantum techniques due to their excellent fluorescent properties. However, synthesis of CDs with controllable spectrum is challenging because of the diversity of the CD components and structures. In this report, machine learning (ML) algorithms were applied to help the synthesis of CDs with predictable photoluminescence (PL) under the excitation wavelengths of 365 and 532 nm. The combination of precursors was used as the variable. The PL peaks of the strongest intensity ( ${\lambda }_{\mathrm{s}}$) and the longest wavelength ( ${\lambda }_{\mathrm{l}}$) were used as target functions. Among six investigated ML models, the random forest (RF) model showed outstanding performance in the prediction of the PL peaks.

All-inorganic lead halide perovskite CsPbX₃ (X = Cl, Br, and I) nanocrystals (NCs) have shown great application prospects in optoelectronic fields. Their properties can be feasibly tuned by the ratio of different halide ions. Post-synthesis halide anion exchange of Cl‒Br or Br‒I in CsPbX₃ NCs allows getting any desired composition of CsPbCl_xBr_3−x and CsPbBr_xI_3−x (0 ≤ x ≤ 3). However, due to the large difference of the Cl and I radii, they can only substitute each other in a limited ratio to form CsPbCl_yI_3−y (0 < y < δ or 3 − δ^' < y < 3). To date, little has been known on the phase diagram of the ternary halide perovskite of CsPbCl _aBr_bI_3−a−b (0 < a + b < 3). In this work, the ternary halide perovskite phase diagram is constructed by the strategy of halide anion exchange between perovskite NCs. From the diagram, the composition and proportion of the perovskite NC final phases from any starting perovskite NC mixture can be calculated. Specifically, a two-phase perovskite NC system showing stable dual photoluminescence (PL) peaks is achieved.