De-noising of images
11th of December 2020
10:00-13:00
Registration is closed as the seminar took place. If you are interested in the material links are provided below and at the GitHub repository.
Note that this workshop will be completely online. The instructor will be online to answer all your questions and guide you through the theory part and the hands-on parts. No in-person presence is possible due to the COVID-19 pandemic.
Agenda
10:00-11:00 - Theory of neural networks and autoencoders applied to the problem of de-noising images
11:00-12:00 - Hands-on session with guided exercises and examples (Python) - Inlcuding lunch while programming
12:00-13:00 - Invited Talk (see below)
Material
The material can be accessed from the GitHub Repository. For more convenience here are the slides and the hands-on Google colab jupyter notebooks
Slides
Slides on Denoising autoencoders
Jupyter Notebooks (Python, the links will open a Google Colab instance, no need to install anything locally)
Invited Talk (speaker will present live)
Title: Physically constrained causal noise models for high-contrast imaging of exoplanets
Speaker: Timothy Gebhard (see Team page)
Abstract:
The detection of exoplanets in high-contrast imaging (HCI) data hinges on post-processing methods to remove spurious light from the host star. So far, existing methods for this task hardly utilize any of the available domain knowledge about the problem explicitly. In this talk, I will discuss a new approach to HCI post-processing based on the half-sibling regression framework from the causality literature. This framework does not only allow us to combine machine learning with existing scientific domain knowledge in a natural fashion, but the resulting de-noising method also performs up to a factor of four times better than one of the currently leading algorithms.
Technical Prerequisites
You will need to have access to a reasonably fast internet connection to be able to follow the live stream of the lecture. The exercises will be all carried out in Google Colab, therefore you only need Google Chrome installed on your computer. A google account is also necessary.
Know-how Prerequisites
To be able to follow this lecture, you will need an intermediate understanding of mathematics, linear algebra and statistics. No previous knowledge of neural networks is assumed. An intermediate Python experience will be required to be able to follow and work on the exercises.
ABSTRACT
In this seminar we will discuss the problem of applying neural networks to de-noise high resolution images. We will discuss some advanced topics about autoencoders and in particular which architectures enable us to deal with 2-dimensional data structures as images. We will use convolutional layers and build autoeconders that are able to deal with images. For example our code will allow us to transform the images as you can see in Figure 1 (source https://www.kaggle.com/anmour/convolutional-autoencoder-with-keras).
Figure 1: What the effect of an autoencoder trained for de-noising can be. Source https://www.kaggle.com/anmour/convolutional-autoencoder-with-keras.