Despite decades of theoretical research, the nature of the self-driven collective motion remains indigestible and controversial, while the phase transition process of its dynamic is a major research issue. Recent methods propose to infer the phase transition process from various artificially extracted features using machine learning. In this thesis, we propose a new order parameter by using machine learning to quantify the synchronization degree of the self-driven collective system from the perspective of the number of clusters. Furthermore, we construct a powerful model based on the graph network to determine the long-term evolution of the self-driven collective system from the initial position of the particles, without any manual features. Results show that this method has strong predictive power, and is suitable for various noises. Our method can provide reference for the research of other physical systems with local interactions.

Enormous progresses to understand the jamming transition have been driven via simulating purely repulsive particles which were somehow idealized in the past two decades. While the attractive systems are both theoretical and practical compared with repulsive systems. By studying the statistics of rigid clusters, we find that the critical packing fraction ϕ c varies linearly with attraction μ for different system sizes when the range of attraction is short. While for systems with long-range attractions, however, the slope of ϕ c appears significantly different, which means that there are two distinct jamming scenarios. In this paper, we focus our main attention on short-range attractions scenario and define a new quantity named “short-range attraction susceptibility” χ p, which describes the degree of response of the probability of finding jammed states p j to short-range attraction strength μ. Our central results are that χ p diverges in the thermodynamic limit as χ p ∝ | ϕ − ϕ c ∞ | − γ p , where ϕ c ∞ is the packing fraction at the jamming transition for the infinite system in the absence of attraction. χ p obeys scaling collapse with a scaling function in both two and three dimensions, illuminating that the jamming transition can be considered as a phase transition as proposed in previous work.

Defining the structure characteristics of amorphous materials is one of the fundamental problems that need to be solved urgently in complex materials because of their complex structure and long-range disorder. In this study, we develop an interpretable deep learning model capable of accurately classifying amorphous configurations and characterizing their structural properties. The results demonstrate that the multi-dimensional hybrid convolutional neural network can classify the 2D liquids and amorphous solids of molecular dynamics simulation. The classification process does not make a priori assumptions on the amorphous particle environment, and the accuracy is 92.75%, which is better than other convolutional neural networks. Moreover, our model utilizes the gradient-weighted activation-like mapping method, which generates activation-like heat maps that can precisely identify important structures in the amorphous configuration maps. We obtain an order parameter from the heatmap and conduct finite scale analysis of this parameter. Our findings demonstrate that the order parameter effectively captures the amorphous phase transition process across various systems. These results hold significant scientific implications for the study of amorphous structural characteristics via deep learning.

In recent years, LegalAI has rapidly attracted the attention of AI researchers and legal professionals alike. Elements of LegalAI are known as legal elements. These elements can bring intermediate supervisory information to the judicial trial task and make the model’s prediction results more interpretable. This paper proposes a Chinese legal element identification method based on BERT’s contextual relationship capture mechanism to identify the elements by measuring the similarity between legal elements and case descriptions. On the China Law Research Cup 2019 Judicial Artificial Intelligence Challenge (CAIL-2019) dataset, the final result improves 4.2 points over the method based on the BERT model but without using similarity metrics. This research method makes full use of the semantic information of text, which is essential in the judicial field of document processing.