Prof. Zhikui Chen, Dalian University of Technology, China
Abstract: The multi-view learning has been attracting increasing research attention in machine learning, since multimodal data are common in practical applications. Data about the same entity are often collected from different sources or channels with different descriptions, called modalities or views. Thus, it is worth exploring faster, optimized and widely-used multimodal learning. Specifically, first of all, a quick-learning multimodal data learning model is proposed, locating a low-dimensional latent subspace where common semantic features are shared by multiple data sets that view-specific features are considered to avoid semantic bias during the process of common feature learning. Furthermore, considering the sample distribution in practical application, the study of multimodal learning is extended to few-shot learning or even zero-shot learning. Context-aware techniques and deep transfer learning are introduced to address data shortages and reuse knowledge, using intra-modal and cross-modal information in heterogeneous data to Inhibit the effect of specificity in few samples, and an orthogonal method that projects cross-modal features and class attributes onto a Hamming space is proposed to learn a discriminative and binary representation of each modality, transfer knowledge from seen classes to unseen classes that have not been appeared in the training data. Finally, the above methods in various real application scenarios are introduced.
Title: The Asymmetric Evolutionary Game Based on Dynamic Payoff
Speech Abstract: In the case of the constant payoff matrix, the evolutionary stability strategy of the asymmetric evolutionary game can only be a pure strategy combination of 0 and 1. What circumstances can a non-zero or one equilibrium become an evolutionary stable strategy? A series of researches have been carried out in this paper.
For the asymmetric evolutionary game model, the proportion of one-sided strategy selection is introduced into the dynamic payoff matrix. In this case, the condition that the equilibrium point of the copied dynamic equation becomes the asymptotic stable point is studied. The potential evolutionary stable strategy combinations of the dynamic system are obtained. Some conditions for the evolutionary stability strategy are obtained and corresponding examples are given.
he proportion of strategy choice between the two sides is introduced into the asymmetric evolutionary game model, and some conclusions are obtained that the equilibrium point becomes an asymptotically stable point. For the equilibrium points (0, 0),(0, 1),(1, 0), and (1, 1), the sufficient and necessary conditions that these points become asymptotically stable strategies are studied. When the payment is a linear function and the equations are assumed to have solutions, the sufficient condition that the non-zero and non-one equilibrium point becomes an asymptotically stable strategy is studied. when the payment is a nonlinear function, that the equations exist solutions and the non-zero (non-one) equilibrium point becomes asymptotically stable strategy is obtained. Finally, some examples are given.
June 18, 2021
June 29, 2021
Registration Deadline: July 1, 2021
Conference Date: July 9-11, 2021
(Office Time 9:00 - 17:30, Time zone: GMT+8; Monday to Friday)
GSRA-WeChat Official Account