About Me

I am a 21-year-old Computer Science graduate, currently pursuing my M.Sc. in Computer Science at ETH Zürich. I gratuated from Ludwig-Maximilians-Universität in Munich with a strong academic background and practical experience in machine learning, data science, and optimization. I am passionate about leveraging technology to solve complex real-world problems in machine intelligence and beyond.

Skills

  • Python, SQL, C, R, Java, Spark, Haskell, Bash
  • Machine Learning & Data Science (Pandas, numpy, tensorflow, sklearn), Big Data, Linux, MS-Office ...
  • Languages: German (native), English (C1 TOEFL), French (B1+)
  • Interests: Skiing, finance, martial arts, business, Start-Up, politics and philosophy

Achievements

Societies: McKinsey Firsthand, BCG Emeralds, ETH Analytics Club

Projects

  • Mitral Valve (MV) Prediction on Echocardiography Data using U-Nets

    The mitral valve (MV) serves as a crucial component of the heart, directing circulation between the left atrium and left ventricle. It consists of two sections, the anterior and posterior leaflets, both connected to a structural ring called the mitral annulus. Under normal conditions, the left atrium contracts during diastole, allowing blood to pass through the open MV into the expanding left ventricle.

    Extracting the MV from echocardiographic footage is a fundamental step in developing artificial intelligence models that can aid healthcare professionals in various tasks such as disease identification, treatment planning, and real-time surgical guidance. This project focuses on the segmentation of the MV in echocardiographic sequences.

    Objective

    The dataset comprises 65 training videos, each containing three manually annotated frames marking the MV, along with a bounding box for localization. The test dataset includes 20 videos, where segmentation must be performed on every frame.

    Training Videos:
    Train Video

    Data Processing

    After resizing the videos, the training data is split into the expert-labeled data (later used for fine-tuning) and the amateur-labeled data. For the expert data, a data loader is used to augment the dataset because the size of the expert dataset is too small.

    Data Processing

    Loss

    Intersection over Union (IoU) Loss:

    $$\text{IoU} = \frac{|A \cap B|}{|A \cup B|}$$

    with the predicted segmentation mask \( A \) and the ground truth mask \( B \).

    The IoU loss is given by:

    $$\text{IoU Loss} = 1 - \text{IoU}$$

    Training

    A custom U-Net model is used for training. The model is first trained using the entire dataset in its original form and later fine-tuned using the augmented expert data.

    1st Training Phase:
    history = valve_model.fit(resized_array_videos, resized_array_masks, epochs=20, steps_per_epoch=80, validation_split=0.1)
    2nd Fine-Tuning Phase:
    # Activate fine tuning
for layer in valve_model.layers[:2]:
	layer.trainable = False

# Fine tune model on augmented expert data
history = valve_model.fit(expert_train_generator, epochs=10, steps_per_epoch=80)

    Results

    The results of the fully trained model on the test set are shown below. The predicted masks are overlayed onto the original (non-labeled) test videos.

    Example 1:
    Output Video 1
    Example 2:
    Output Video 2

  • Heart-Data Timeseries Analysis

    Analyzed heart data using machine learning techniques to predict anomalies and trends.

  • Satellite Image Classification

    Developed a CNN model to classify satellite images for environmental monitoring.

  • Novel Anomaly Detection Algorithm

    Created an unsupervised algorithm for high-dimensional datasets, achieving benchmark results.

Education

  • ETH Zürich (M.Sc. in Computer Science)

    Specialization: Machine Intelligence (09/2024 - 02/2026)

    Key Interests: Data Science, Mathematics, Machine Learning

  • Ludwig-Maximilians-Universität München (B.Sc. in Computer Science)

    Minor in Business Administration (10/2021 - 08/2024)

    Thesis: Creating a novel unsupervised learning algorithm

  • Staffelsee-Gymnasium Murnau

    Abitur with honours

Experience

  • Advanced Machine Learning Projects (ETH Zürich)

    Worked on time series analysis, brain age prediction, Satellite image classification leveraging state of the art CNN techniques, Medical image segmentation for Heart MRI data utilizing U-nets

  • Research in Unsupervised Anomaly Detection (LMU Munich)

    Developed a novel anomaly detection algorithm for high-dimensional datasets, achieving competitive results in industry standart benchmarks like ADBench.

  • Teaching Assistant (LMU Munich)

    Taught courses including Theoretical Computer Science, Database Systems, and Algorithms and Data Structures, managing large cohorts. Teaching classes, responsible for grading the final exam, creating bonus sheets for up to 800 students.

Contact Me