Search results for: Mingxia Zhu

Authors: Min Jindong, Wang Mingxia

Abstract:

With the widespread use of artificial intelligence in life, computer vision, especially deep convolutional neural network models, has developed rapidly. With the increase of the complexity of the real visual target detection task and the improvement of the recognition accuracy, the target detection network model is also very large. The huge deep neural network model is not conducive to deployment on edge devices with limited resources, and the timeliness of network model inference is poor. In this paper, knowledge distillation is used to compress the huge and complex deep neural network model, and the knowledge contained in the complex network model is comprehensively transferred to another lightweight network model. Different from traditional knowledge distillation methods, we propose a novel knowledge distillation that incorporates multi-faceted features, called M-KD. In this paper, when training and optimizing the deep neural network model for target detection, the knowledge of the soft target output of the teacher network in knowledge distillation, the relationship between the layers of the teacher network and the feature attention map of the hidden layer of the teacher network are transferred to the student network as all knowledge. in the model. At the same time, we also introduce an intermediate transition layer, that is, an intermediate guidance layer, between the teacher network and the student network to make up for the huge difference between the teacher network and the student network. Finally, this paper adds an exploration module to the traditional knowledge distillation teacher-student network model. The student network model not only inherits the knowledge of the teacher network but also explores some new knowledge and characteristics. Comprehensive experiments in this paper using different distillation parameter configurations across multiple datasets and convolutional neural network models demonstrate that our proposed new network model achieves substantial improvements in speed and accuracy performance.

Keywords: object detection, knowledge distillation, convolutional network, model compression

Procedia PDF Downloads 278

Authors: Mingxia Zhu, Matthew Tingchi Liu

Abstract:

AI grows gradually in the market with its efficiency and accuracy, influencing people’s perceptions, attitude, and even consequential behaviors. Current study extends prior research by focusing on AI’s impact on consumers’ morality. First, study 1 tested individuals’ believes about AI and human’s moral perceptions and people’s attribution of moral worth to AI and human. Moral perception refers to a computational system an entity maintains to detect and identify moral violations, while moral worth here denotes whether individual regard an entity as worthy of moral treatment. To identify the effect of AI on consumers’ morality, two studies were employed. Study 1 is a within-subjects survey, while study 2 is an experimental study. In the study 1, one hundred and forty participants were recruited through online survey company in China (M_age = 27.31 years, SD = 7.12 years; 65% female). The participants were asked to assign moral perception and moral worth to AI and human. A paired samples t-test reveals that people generally regard that human has higher moral perception (M_Human = 6.03, SD = .86) than AI (M_AI = 2.79, SD = 1.19; t(139) = 27.07, p < .001; Cohen’s d = 1.41). In addition, another paired samples t-test results showed that people attributed higher moral worth to the human personnel (M_Human = 6.39, SD = .56) compared with AIs (M_AI = 5.43, SD = .85; t(139) = 12.96, p < .001; d = .88). In the next study, two hundred valid samples were recruited from survey company in China (M_age = 27.87 years, SD = 6.68 years; 55% female) and the participants were randomly assigned to two conditions (AI vs. human). After viewing the stimuli of human versus AI, participants are informed that one insurance company would determine the price purely based on their declaration. Therefore, their open-ended answers were coded into ethical, honest behavior and unethical, dishonest behavior according to the design of prior literature. A Chi-square analysis revealed that 64% of the participants would immorally lie towards AI insurance inspector while 42% of participants reported deliberately lower mileage facing with human inspector (χ^2 (1) = 9.71, p = .002). Similarly, the logistic regression results suggested that people would significantly more likely to report fraudulent answer when facing with AI (β = .89, odds ratio = 2.45, Wald = 9.56, p = .002). It is demonstrated that people would be more likely to behave unethically in front of non-human agents, such as AI agent, rather than human. The research findings shed light on new practical ethical issues in human-AI interaction and address the important role of human employees during the process of service delivery in the new era of AI.

Keywords: AI agent, consumer morality, ethical behavior, human-AI interaction

Procedia PDF Downloads 82