14405

بررسى روشهاى بصرى سازى داده هاى غير قطعى و چالش هاى ان

1400

د. محمد رضا كنكاورى

د. خنجري

In this semenar, we introduce SurroFlow, a novel surrogate model for efficient simulation parameter recommendation and exploration. SurroFlow effectively captures the conditional distribution of simulation outcomes conditioned on the simulation parameters, facilitating efficient data generation for different simulation parameters. The complex conditional distribution is modeled by applying a series of invertible and learnable transformations on a base Gaussian distribution. The variance in the Gaussian distribution reflects the uncertainty in the surrogate modeling process. Remarkably, these transformations also enable reverse computation, allowing for the prediction of simulation parameters from simulation data. We integrate SurroFlow with a genetic algorithm and utilize them as the backend for a visual interface, where scientists can interactively specify their optimization objectives through the visual system. After that, the optimization process starts automatically for scientists’ preference-guided parameter exploration. Our quantitative and qualitative results demonstrate SurroFlow’s ability to predict high-fidelity simulation outputs, navigate simulation parameter spaces efficiently, and quantify uncertainties.

12/11/2024 12:00:00 AM

85563

1403/10/29

The study of Uncertain Data Visualization Techniques and its challenges

(عدم قطعيت داده ها، تكنيك هاي تجسم، نمايش خطا، تصميم گيري، عدم قطعيت چند بعدي و رويكردهاي تجسم تركيبي.)

(Data Uncertainty, Visualization Techniques, Error Representation, Decision-Making, Multidimensional Uncertainty, and Hybrid Visualization Approaches.)