I am a master student in Astronomy and Data Science at Leiden University. I am mainly interested in applying various data science techniques to the big data that modern astronomy is producing. My master's programme is the perfect blend of these topics, with multiple courses on data mining and deep-learning, such as `Advances in Data mining' and `Neural networks' as well as general Astronomy courses such as `Star and Planet formation' and `Origin and evolution of the Universe'. Next to my studies, I really enjoy helping fellow astronomy students. I have given seminars for the course 'Introduction to Astrophysics' and have been a tutor for 4 years now, which entails helping a group of about 5 students get through their first year. Additionally, I like to go to the gym a couple times per week to fight gravity.
I am hoping to pursue a PhD programme where I can combine all my interests: data science, astronomy and teaching.
Feel free to have a look around this website that functions like an interactive informal CV or check out my CV directly
Over the next decade, "Stage IV" lensing surveys such as Euclid will revolutionize our view of the Universe by providing some of the tightest constraints on cosmological parameters. To accomplish this, these surveys will detect billions of galaxies. A crucial step in the inference of cosmological parameters is compressing this large volume of data to a manageable amount of statistical summaries. In this work, I use a neural network to find an optimal compression function for a Euclid-like survey while preserving physical information.
Work in progress
The far infrared regime between 15 and 1000 micron is ideal for observations of cold gas and dust. However, observations in this regime can almost exclusively be performed by satellite based observatories with relatively small mirror diameters, which causes very limited resolution. To improve the resolution of these observations, a novel neural network based approach was explored. A convolutional neural network was trained to predict high resolution images. This was achieved by combining high resolution observations in the optical and near-infrared by Earth-based observatories like Subaru and VISTA together with lower resolution FIR data.
A picture of the network architecture we used is given below.
In my first master thesis I tested the hypothesis that the position angles of radio galaxies are randomly oriented in the HETDEX Spring Field region. I extracted a sample of 6795 double-lobed radio galaxies from the sample of 325,694 low frequency radio sources in the first data release of the LOFAR Two-metre Sky Survey.
I performed two statistical tests on the sample, one in two dimensions and one in three dimensions with the 3416 sources that have a redshift measurement available. After taking into consideration possible uncertainties in the position angles, I found that a the subset of highest flux sources showed a significant (p-value < 0.001) deviation from uniformity at angular scales between 4 and 5 degrees in the 2D analysis.
The significance level is plotted as a function of angular scale for four increasing flux bins below:
While there were multiple indications that this result might be caused by biases in the dataset, the result could not be ruled out with certainty.
Supposing that the effect is not caused by systematic effects, the alignment of the active galactic nuclei that give rise to these radio jets suggests large scale coherence in angular momentum, which would reinforce prior evidence for the alignment of active galactic nuclei with their surrounding large scale structure inferred from polarization alignment studies.
In the astronomy bachelor at Leiden University, the bachelor research project is done in groups of two. We used Python to analyze spectroscopic data taken of M20 with the MUSE instrument at the VLT. We created maps of the dust extinction, electron temperature and velocity along the line of sight of the whole nebula. Two structures were investigated more thoroughly. The velocity and mass loss for a Herbig Haro jet (H399) was estimated and the nature of a bow shock-like structure was investigated.
A map of the density of M20 in electrons per cubic centimeter is given below. It was calculated using the two temperature sensitive line ratios: [NII]5755/6548 and [NII]5755/6583 and the density sensitive line ratio [SII]6731/6716
Gave a talk about my first master's project at the LOFAR Key Science project meeting, which is a meeting of the international LOFAR consortium. See the PDF here
See also: Research Experience
Includes a minor in Data Science, which focuses on learning the basics of data analysis and pattern recognition in big data.
For the course Business Intelligence and Process Modelling in the Data Science minor we designed a dashboard. The idea was to design a website where the sales of the mobile application "CookieDestroyer" could be interactively viewed. Check it out here!
For the course Neural Networks we trained several neural networks to predict road networks based on satellite images. This was done using data provided by the Google Maps API. The network was able to find the large features in the images, but the roads were still too difficult to learn..