Data-driven machine learning models for the quick and accurate prediction of Tg and Td of OLED materials

Abstract

Organic light-emitting-diode (OLED) materials have exhibited a wide range of applications. However, further development and commercialization of OLEDs requires higher-quality OLED materials, including high thermal stability associated with the glass transition temperature (Tg) and decomposition temperature (Td). Experimental determinations of the two important properties genernally involve a time-consuming and laborious process. Thus, it is highly desired to develop a quick and accurate prediction tool. Motivated by the changelle, we explored machine learning by constructing new dataset with more than one thousand samples collected from a wide range of literaturesm, through which ensemble learning models were explored. Models trained with the LightGBM algorithm exhibit the best prediction performance, where the values of MAE, RMSE, and R 2 are 17.15 K, 24.63 K, and 0.77 for Tg prediction, 24.91 K, 33.88 K, and 0.78 for Td prediction. The prediction performance and the generalization of the machine learning models are further tested by out-of-sample dataset, also exhibiting satisfactory results. Experimental verification further demonstrates the reliability and the practical potential of the ML-based model. In order to extend the practice application of the ML-based models, an online prediction platform was constructed, including the optimal predition models and all the thermal stability data under study, which are freely available at http://oledtppxmpugroup.com.We expect that they will become a useful tool for experimental investigations on Tg and Td, in turn accelerating the design of the OLED materials with high performance.

Towards trustworthy multi‐modal motion prediction: Holistic evaluation and interpretability of outputs

“…Recently, many different approaches were proposed to deal with multi-modal motion prediction [12][13][14][15][16][17]. The state-of-theart methods rely on graph models [10,11,[18][19][20], since traffic scenarios can be naturally represented as graphs. The graph structure is flexible and efficient because it allows direct and explicit representation of interactions through edges.…”

“…Alternatively, less expensive graph-based representation can be used in the form of polylines (e.g., lanes, crosswalks, and boundaries), which represent piecewise linear segments [10,16,18,19,23,24]. In this representation, long-range dependencies are effectively and efficiently modelled.…”

“…In this representation, long-range dependencies are effectively and efficiently modelled. For example, Vector-Net [10,18,19] represents the lane segments and agents histories as vectors connected end-to-end. As an improvement, the Lane-based Trajectory Prediction (LTP) method [23] encodes the map through connected sliced lane segments to more precisely model the intention.…”

“…Based on the polyline representation, VectorNet [10,18,19] utilises only self-attention modules to directly learn the interactions between all the sub-graphs in the environment. Further, goal-oriented lane attention [31] emphasises the relationship between agents and lanes.…”

“…On the other hand, the proposal-based methods design various possible proposals and thus separately solve the tasks of intention prediction and motion prediction. Some methods [10,19] sample proposal points around the lane centreline to capture detailed information. The proposed sampling is based on carefully developed rules, the restriction on sharing proposals between different agents is one of them.…”