Scroll to Section:

Exoplanets are planets beyond our own solar system. Since they do not emit much light and moreover are very close to their parent stars they are difficult to detect directly. When searching for exoplanets, astronomers use telescopes to monitor the brightness of the parent star under investigation: Changes in brightness can point to a passing planet that obstructs part of the star’s surface. The recorded signal, however, contains not only the physical signal of the star but also systematic errors caused by the instrument. As BERNHARD SCHÖLKOPF explains in this video, this noise can be removed by comparing the signal of the star of interest to those of a large number of other stars. Commonalities in their signals might be due to confounding effects of the instrument. Using machine learning, these observations can be used to train a system to predict the errors and correct the light curves.
DOI:
https://doi.org/10.21036/LTPUB10261

Researcher

Bernhard Schölkopf is Director of the Max Planck Institute for Intelligent Systems in Tübingen and the head of the Department for Empirical Inference. He studied physics, mathematics, and philosophy in both Tübingen (Germany) and London. Since 2002 he is an Honorary Professor at Technical University Berlin.

Schölkopf’s main research interest concerns inference from empirical data and machine learning. He applies his research on machine learning to the exploration of exoplanets, that is planets beyond our solar system.

Institution

Max Planck Institute for Intelligent Systems

The Max Planck Institute for Intelligent Systems is a world leading research institution in the field of AI, Machine Learning and Robotics. Our goal is to understand and investigate the fundamental problems of Perception, Action and Learning, asking the question: How are intelligent systems able to operate autonomously in, and adapt to, complex, changing environments? We use nature’s blueprint to design state-of-the-art artificially intelligent systems. They range from life-like avatars to humanoid robots with a sense of touch, to algorithms that detect patterns in huge datasets.
  
Spanning two campuses in Stuttgart and in Tübingen, MPI-IS is located in the heart of the State of Baden-Württemberg, one of the leading economic regions in Europe. Together with its neighboring institutes, Stuttgart and Tübingen form one of the biggest clusters within the Max Planck Society – Germany’s most successful research organization and one of the leading institutions for basic research worldwide.
 
The Stuttgart campus concentrates on physical realizations of intelligent systems. It has world leading expertise in mobile micro robotics, haptics, and soft and bioinspired robotics. One focus is creating wireless tiny medical robots that could one day operate inside our body and by that revolutionize medicine and healthcare. We also study physical human-robot interactions and touch sensing for it to play a more prominent role in robotics. We develop innovative robotic materials – the building components of tomorrow’s robots, we develop bio-inspired systems and much more.
 
The Tübingen site focuses on the computational aspects of intelligence, with departments and research groups in the fields of computer vision, machine learning, algorithms, theory and robotics. We also conduct research on the social foundations of computation. Our researchers capture the shape and movement of the human body to create realistic avatars, they teach robots to learn autonomously how to walk and how to interact with the environment, and much more.
 
The Max Planck Institute for Intelligent Systems thrives thanks to the many highly motivated and talented people that work here. Its international and diverse community brings together the most curious doctoral researchers, outstanding aspiring scientists, and established leaders in their field. Together, they push the limits of tomorrow’s state of the art of artificially intelligent systems.

Show more

Original publication

Removing Systematic Errors for Exoplanet Search via Latent Causes

Schölkopf Bernhard, Hogg David W., Wang Dun, Foreman-Mackey Daniel, Janzing Dominik, Simon-Gabriel Carl-Johann and Peters Jonas
arXiv preprint arXiv:1505.03036
Published in 2015

Reading recommendations

Modeling Confounding by Half-Sibling Regression

Schölkopf Bernhard, Hogg David W., Wang Dun, Foreman-Mackey Daniel, Janzing Dominik, Simon-Gabriel Carl-Johann and Peters Jonas
Proceedings of the National Academy of Sciences
Published in 2016

Learning with Kernels

Schölkopf Bernhard and Alexander J. Smola
Published in 2002

A Systematic Search for Transiting Planets in the K2 Data

Schölkopf Bernhard, Hogg David W., Wang Dun, Foreman-Mackey Daniel, Montet Benjamin T. and Morton Timothy D.
The Astrophysical Journal
Published in 2015

Beyond