Contrastive learning, a self-supervised learning method, has become one of the main techniques for visual representation learning. It builds contrastive views through data augmentation, maximizing the mutual information between views with the same semantic information. Currently, the main methods used for data augmentation are random cropping, resizing, rotating, and recoloring images. However, due to the diversity and randomness of data augmentation, it is difficult to guarantee the same semantic information between pairs of positive examples during data augmentation. This limits the efficiency and applicability of contrastive learning. To circumvent this problem, in this paper, we propose a Simple Contrastive Learning(SimCL) method for image classification, which abandons data augmentation and constructs contrastive views by adding uniform noise in the representation space for contrastive learning. Specifically, the original image is passed through a convolutional neural network to obtain a representation of the image, and a contrastive view is constructed by adding different noises to the representation. Despite its apparent simplicity, comprehensive experiments on 11 benchmark datasets show that the method can match or even outperform data augmentation-based contrastive learning methods.

This paper investigates the dissipativity analysis of large-scale networked systems with linear time-invariant dynamics. The networked system is composed of a large number of subsystems whose connections are arbitrary, and each subsystem can have different dynamics. A sufficient and necessary condition for the strict dissipativity analysis of the networked system is derived, which takes advantage of the block-diagonal structure of the system parameter matrix and the sparseness characteristics of the subsystem interconnections. Then, a necessary condition and a sufficient condition that depend only on a single subsystem parameter are given separately. Numerical simulations illustrate that compared with the existing results, the conditions suggested in this paper have higher computational efficiency in the dissipative analysis of large-scale networked systems.

The controllability of multiagent systems governed by node and edge dynamics on signed graphs is investigated under distinct partition rules and control algorithms. Motivated by a social network case study, dynamic models for nodes and edges are established. A novel concept “edge bidirectional mapping” is introduced, along with an associated matrix that rigorously characterizes the relationship between node and edge Laplacian matrices. Necessary and sufficient conditions for controllability equivalence between node and edge dynamics are derived across three topological regimes, determined by the cardinality relationships between the number of nodes n and edges m, namely n > m, n = m, and n < m. Furthermore, more explicit and comprehensive criteria for controllability properties of node and edge dynamics are formulated under distinct control algorithms and partition rules. Specifically, the proposed algorithms are considered in both absolute and relative forms, while the partition rules, distance partition (DP), generalized almost equitable partition (GAEP), and generalized relaxed almost equitable partition (GRAEP), are distinguished based on the number of positive, negative, and total connections between vertices (nodes or edges) within a cell and their neighboring connections to other cells. Notably, this study systematically analyzes their hierarchical inclusion relationships for the first time. Upper and lower bounds on the dimensions of the node and edge controllable subspaces are presented. The lower bound reflects DP with edge weight considerations, while the upper bound is obtained via partition algorithms that yield the best LGRAEP. Additionally, the computational complexity and convergence properties of the proposed partition rules (DP, GEP, GREP), Algorithms 1-2, and the derived bounds for the controllable subspace dimensions are analyzed. Theoretical results are comprehensively validated through multi-level simulations, including verification of theoretical bounds, analysis of clustering effects, and comparative model evaluation.