The Art of Inference: Practicing Bayesian Reasoning in Computer Vision Problems


Bayesian reasoning has a long standing history than frequentist since 18th century. In this talk, I'd like to present those fancy techniques appeared in many forensic movies and TV shows. Including the restorations of noisy and scratched images, deblur the unfocoused pictures...etc. These topics are no-doubt very active in the domain of computational photography.

I will perform plenty of vivid examples on how to achieve these goals by Bayesian reasoning and Monte-Carlo simulations together with python practicing. Alternatively, such goals can be accomplished by modern convolutional neural network (CNN) as well. I will conduct a short comparison between the traditional and modern ones.

The audience does not require to have any specific knowledge on probability theory though basic calculus will be helpful.


- [cv2] OpenCV for python - [keras] Will be used to build the CNN model that has a similar capability to the traditional computational photography algorithms - [scipy] Efficient scientific computational package for python



Yen-Hsun Lin

I'm a theoretical physicist that is enthusiastic in revealing the mystery of our resplendent Universe. Though people often considered math is a daunting task, I always found harmony and elegance behind it. Despite the constant debate on how should we proceed the understanding of nature epistemologically or ontologically, I believe that the essence of science lies in humanity itself instead of such philosophical conundrum.

My major research includes dark matter detection, astroparticle physics and computer simulations with some side projects on computational photography.

I will join the Institute of Physics, Academia Sinica as a Distinguished Postdoctoral Scholar this August unless surprise happens.