Cognitive Model for Probability Density Distribution Uncertainty Visualization

Abstract: People often obtain information and make judgments with the help of systems and machines, but due to the uncertainty in the instrument, environment, and data processing, the judgments made by the system are not always correct. Blindly believing the results of the system's decision-making is very dangerous, so effective information exchange between the system and users is the prerequisite to ensure correct decision-making, and it is extremely important to convey the uncertain information of the system to users. Visualization is an important way to inform users of uncertain information. At present, many scholars have conducted research on how to express the uncertainty contained in data. Then whether users can effectively recognize and understand these visual graphics becomes an issues to be explored.What this article wants to explore is how people recognize and understand visualization for uncertainty. Only by understanding the way people interpret visualization can the visualization be optimized and the information exchange between the system and users can be better realized.In this paper, the user’s perception of probability under the triangular distribution is studied. A symmetric distribution model and two asymmetric models are used. In each distribution model, 13 test points are selected according to the true value of the probability density. The experiment is realized by Unity3D. In each trial, the positions of the distribution model and the test points appear randomly. The experiment describes the task of using remote sensing technology to recognize camouflage targets. The participants need to answer the probability that the target to be detected represented by the test point is the enemy's camouflage target.This paper verifies the effectiveness of using the brightness gradient to visualize the uncertainty, and this conclusion is consistent with the results of previous studies. At the same time, this paper proposes a fitting model of the user's internal probabilistic cognition and the visual expression of probability density. We found that when the participants used the triangular probability distribution information expressed by the brightness gradient to infer the probability that the test point is a camouflage target, this perceived probability has a strong linear correlation with the true value of the probability density corresponding to the point. Participants will give a probability prediction close to 1 at the test point with the lowest brightness, that is, the highest probability density, and a probability prediction close to 0 at the highest brightness point, that is, the lowest probability density point. Fitting based on the linear model can obtain the probabilistic cognitive model of the subjects for the visualized graph.This article provides a way to realize the communication of uncertainty information between the system and the user, and may contribute some help to the effectiveness and reliability of artificial intelligence.

Keywords: Uncertainty Visualization, Cognition, Remote Sensing, Target Recognition

DOI: 10.54941/ahfe100933

Cite this paper:

​