Machine Learning: The elegant way to extract information from data
Drs Jens Kauffmann, Veselina Kalinova, Dario Colombo and Tobias Albertson (MPIfR)
The human brain is able to recognize patterns in the data due to the millions of years of evolution. Today, we try to use and further develop computer algorithms that can reproduce this ability to distinguish patterns in large statistical samples and images, to find correlations between events, to classify objects based on their similar and different features in an objective and automatic way. Machine Learning is a sub-field in computer science that explores pattern recognition in the data analysis and characterize already known trends in sets of observations. This course focuses on the basic most applied up-to-date techniques in the machine learning that can be used in any branch of science and industry. We start with a broad introduction about the topic, followed by detailed discussion about the theory and application of the different types of the machine learning — supervised and unsupervised learning. In the field of supervised learning, for example, we will discuss regression, support vector machines, and neural networks, while the field of unsupervised learning we will cover clustering techniques, principal component analysis, and dimensionality reduction. The final goal of the course is to give the hands–on knowledge needed to pick and apply machine learning tools in Python. Discussions of the underlying mathematical principles will illustrate the inner workings of these tools.
Topics to be covered
Introduction — what is machine learning?
Supervised Learning — regression, support vector machines, neural networks
Unsupervised Learning — clustering, principal component analysis, dimensionality reduction
Tools — Monte Carlo Markov chains, Bayesian inference
Dynamics of star clusters containing stellar mass black holes
Prof. Hyung Mok Lee (Seoul National University)
LIGO has detected three gravitational wave events due to the merger of black hole binaries. The black hole binaries could have formed through the evolution of binaries composed of massive stars. However, such binaries could have been formed in globular clusters purely by the dynamical processes because of high stellar density. Also globular clusters are very old objects with low metallicity. Among three black hole binaries, two are known to have been composed of relatively massive ones. Such a massive black holes can be formed in low metallicity environment. In the presence of massive components such as black holes, the mass segregation could take place rapidly, forming very dense core purely composed of black holes. Binaries of black holes can form either three-body or two-body processes. Under typical physical conditions of the central parts of the globular clusters, three-body processes are thought to be more efficient. The binaries formed via three-body processes are relatively wide and the time scale for the binaries to merge would be much longer than Hubble time. However, the close interactions between binaries and single black holes make the orbits tighter. Such a tightening of binary orbit continues until the binaries get ejected since the recoil energy per interaction increases as the orbit becomes smaller. Some fraction of the ejected binaries eventually will undergo merger within Hubble time. This lecture intends to provide an overview of the gravitational waves, dynamical evolution of dense stellar systems, formation processes of binaries in high density regions, and the interaction between binaries and singles.
Topics to be covered:
- Introduction - Gravitational Waves
- Dynamical evolution driven by two-body relaxation
- Methods of stellar dynamics
- Formation Processes of Binary Stars
- Close interaction of stars with binaries
Link for registration: https://events.mpifr-bonn.mpg.de/indico/event/51/