Simon K., Python Software Engineer with data engineering skills

Experience

Deep Learning Intern – Multi-task learning, Bosch Center for Artificial Intelligence (BCAI)

03/2022 – 09/2022 (Renningen, Germany)

• Developed novel loss weighting methods for Multi-task learning that outer-formed state-of-the-art methods
• Compared loss balancing methods from the literature on tasks such as semantic segmentation, depth, and normal surface estimation on scene understanding datasets such as Cityscapes and NYUv2
• Registered two novel loss weighting methods as patents
• Documented the results within the master thesis

Technologies:  Python, PyTorch, MTL, Git, IBM Spectrum LSF

Technical Solutions Specialist/Data Engineer, Scalefree International GmbH

10/2020 – 01/2022 (Hanover, Germany)

• Led the internal BI development team as a Scrum Master
• Established connection between different source systems and the enterprise data warehouse
• Developed processes for loading the staging area and the raw data vault by employing AWS services such as S3, Lambda, and Batch
• Created XML documents using T-SQL and XQuery for an external customer project
• Containerized jobs using docker and yaml to load the enterprise data warehouse and deployed them using AWS batch.

Technologies:  SQL, Python, Linux, Git, Bash Script, Data, Vault, AWS, YAML

Machine Learning Intern – Cloud ML Services, Novatec Consulting GmbH

11/2019 – 06/2020 (Hanover, Germany)

• Developed a prototype application for churn prediction and evaluating different machine learning algorithms such as Random Forest, SVM, Gradient Boosted Decision Trees, and Logistic Regression using sci-kit-learn
• Compared the Cloud Machine Learning Services MS Azure Databricks with PySpark, AWS Sagemaker, and Google Cloud BigQuery
• Documented the results of the comparison within the bachelor thesis Python Scikit-learn

Technologies: PySpark, MS Azure, AWS GCP

Academic Projects

Student Research Project, University of Hildesheim

12/2020 – 03/2022 (Hildesheim, Germany)

• Conducted image-to-image translation between the domains of regular images and artworks with Deep Generative Adversarial Networks using Tensor-Flow
• Enhanced CycleGAN by introducing a two-objective discriminator as regularization, incorporating adversarial self-defense for better cycle consistency, and applying differentiable augmentation on the target domain with fewer data
• Employed agile intercultural project management techniques to manage the project successfully

Technologies: Python, TensorFlow, GANs, Git, Slurm

Coursework

Machine Learning, University of Hildesheim

04/2020 – 09/2021 (Hildesheim, Germany)

• Implemented various machine learning models such as ridge regression with SGD, LASSO with coordinate descent, least-angle regression, logistic regression with Newton method, gradient-boosted decision tree, and AdaBoost from scratch in Python and NumPy on real-world datasets like Rossmann sales and Wine quality data. Employed data preprocessing techniques such as one-hot encoding, stratified sampling, PCA, and KNN data imputation
• Conducted performance comparison of the implemented models with a sci-kit-learn implementation
• Performed exploratory data analysis on various real-world datasets using Pandas and Matplotlib
• Developed a recommender system by applying matrix factorization with SGD on a movie lens 100k dataset

Technologies:  Python, NumPy, Pandas, sci-kit-learn, Matplotlib

Deep Learning/Computer Vision, University of Hildesheim

04/2021 – 09/2021 (Hildesheim, Germany)

• Trained a CNN end-to-end on a self-driving dataset (camera view from the car) using regularization techniques such as cutout and mixup and implemented a custom batch normalization layer and residual connections to predict the steering angle in PyTorch
• Computed the saliency map for an input image using an ImageNet pre-trained model
• Compared metric learning techniques such as learned embedding of a simple classification model, contrastive loss, and triplet loss with an embedding layer for MNIST data using TensorFlow
• Implemented transfer learning for training a U-Net model on a real-world weed field image dataset with a custom categorical cross-entropy loss. Pretrained the first half of the model on the classification dataset DeepWeeds using TensorFlow, improving the test accuracy by 1.5% compared to a vanilla U-net model, and visualized the predicted segmentation map
• Generated adversarial examples using the Carlini-Wagner attack against a CNN trained on MNIST data and created sparse perturbations with the Hoyer-Square regularizer using PyTorch

Technologies: Python, PyTorch, TensorFlow

Distributed Computing, University of Hildesheim

04/2020 – 03/2021 (Hildesheim, Germany)

• Performed exploratory data analysis using PySpark on the movie lens 10m dataset and used the Hadoop MapReduce framework on BTS flight data
• Conducted distributed K-means clustering and distributed linear regression using SGD on KDD Cup 1998 dataset and VirusShare executables with OpenMPI, including a performance analysis on the speed-up with different numbers of used cores
• Implemented Naive Bayes and SVM classifiers from scratch to categorize news items on 20 newsgroups text datasets using preprocessing techniques such as bag-of-words and TF-IDF feature representation and the Hadoop MapReduce framework
• Employed distributed matrix factorization using coordinate descent with the Hadoop MapReduce framework on the movie lens 10m dataset

Technologies: Python, Hadoop, MapReduce, PySpark, OpenMPI, mpi4py

Reinforcement Learning, University of Hildesheim

10/2022 – 03/2023 (Hildesheim, Germany)

• Utilized PyTorch to develop both the Deep Q-Learning model and the REINFORCE algorithm with policy gradients from scratch to solve the Gym environment Mountain Car

Technologies: Python, PyTorch

Education

M.S. Data Analytics, University of Hildesheim

04/2020 – 01/2023 Hildesheim, Germany
GPA: 3.5/4.0

B.S. Business Information Systems, University of Applied Sciences and Arts Hanover

03/2016 – 06/2020 Hanover, Germany
GPA: 3.5/4.0

Certificates

• Certified Data Vault 2.0 Practitioner (CDVP2)
• Professional Scrum Master (PSM I)