Maximilian Sackel
Welcome to my personal side. Scroll to get more information.
A physis on his way to discover the world.
In my free time I like to experience adventures with friends and learn new things. This is not limited to one specific area and I'm always open to something new to venture.
Since my early childhood I am practicing judo. Judo can be loosely translated as "winning by yielding". This means that to achieve your goals, you have to study your opponent and try to use the force that the opponent invests against himself, with a tiny bit of your own force to overcome your own weakness. This was one of my first lessons that the use of physical principles can make the difference.
Fields of Interest and Experties
![](images/shower.jpg)
Monte Carlo Simulations
In cases data are difficult to obtain, theories are available or statistical systems are observed, the study of Monte Carlo simulations can give evidence about certain circumstances.
One application area for Monte Carlo simulations are extended air showers, which I investigated during my studies. If a astrophysical particle hits the atmosphere, it can partitionate parts of it's primary energy into the production of secondaries particle over a stochastical process. These secondaries particles can produce further particles if there energy is above a threshold.
Particle cascades in astophysical context can be simulated with the CORSIKA library. It can be used to measure the signal a primary particle produce for different detection channels. In my master thesis I implemented the first electromagnetic interaction model in CORSIKA's current version. Therefore I used the lepton and photon propagator PROPOSAL to simulate the production of secondaries particle and implement a module which communicates with CORSIKA.
Big Data Analysis
For a existing dataset the task is to generate the maximum information gain according to a problem definition. Very large amounts of data have to be processed and an uncertainty on the predictions of the models has to be estimated. Deep Learning offers a possibility to solve both problems in an appropriate way.
To make a decision based on data, a problem must be defined so that the data can be narrowed down to problem area. Based on the problem area models can be trained which search for outliers or make predictions. The models which exhibit a good generalization behavior are searched and optimized. In order to make robust statements, uncertainties are applied to the predictions in order to fall back on additional decision factors in case of doubt.
During my studies I was able to gain experience in the application of machine learning in several projects, such as direction reconstruction or type classification of particles. In my spare time I am working on projects like automatic classification of cloud types or football prediction.
![](images/deep_learning.jpg)
![](images/software_development.jpg)
Software Development
Why not doing it right from the beginning on? In order to avoid doing work more than once and to avoid mistakes I try to constantly develop my package operated development.
Complex problems are easier to solve if they are breaked to multiple handable problems. For that I try to improve the package development skills in Python and C++ and solve the problems by modularization. Structural and programming errors are minimized by test-driven development and early documentation keeps the project entry barrier low.