Projects
Predict the Dangerous Places of Car Accidents
We finished this project with prediction-based optimization methods. First, with recent machine learning technology, we predict a probability that car accidents occur in each place. With the predicted danger, we find the place where things for traffic safety are the most effective, considering the effect of a safety thing can be reduced when it is installed near other safety things.
Web Program for Crowd Sourcing [git]
In the DB courses, we made the web program based on a database system. This program receives CSV files from users, make reviewers evaluate the files, and save the file in the database with selected forms.
My Idea (non-published)
GANs with ODEs and Metric learning [git]
This is the extended version of OCT-GAN. We think that it is hard for the simple classifier of the discriminator to fully distinguish fake and real paths with GAN losses. Therefore, we add the metric model which can find the discrepancy between real and fake paths with metric losses. Because metric losses are different from GAN losses, the path discrepancies that the metric model and the classifier find are different. Therefore, using QA-GAN which can consider both GAN losses and another metric, the metric model can assist the classifier in the discriminator.