Mahmoud Aslan Email & Phone Number

Munich, Bavaria, Germany

Budapest, Hungary

- Working in the Neural Information Processing Group (NIPG) led by András Lörincz.- Contributed to a Python codebase (spanning 4k lines of code) to support three scientific ML publications.- Implemented precise optimization algorithms using Python in the field of Structured and Convolutional Sparse Coding, Iterative Optimization, Dictionary Learning, and Adversarial Robustness in Object Recognition and Computer Vision.- Refactored and integrated multiple Python repositories and conducted over 500 experiments on various datasets using Pytorch, Tensorboard, and Jupyter notebooks studying the adversarial robustness of different sparse coding methods.- Implemented FISTA based sparse fully-connected and convolutional neural networks, testing different architectures (ResNet, Wide-ResNet, VGG, etc) and state-of-the-art training techniques (adversarial, weight pruning), studying their behavior under various regularizations, while bridging the gap to theory.- Engineered model training, tracking, and persistence using industry-standard and in-house solutions.- Performed extensive empirical studies on the adversarial robustness of different sparse coding methods (sparse, group sparse, and pooled group sparse representations) in a Linux environment utilizing Pytorch, Tensorboard, Jupyter Notebooks, Singularity, Docker, Bitbucket, Azure VMs, and on-premises GPUs.- Produced empirical evidence that the group representations can be used as a fully unsupervised classification method.- Published the results in DeLTA 2022 and contributed the implementations to the group’s codebase using Pytorch and Python.

Jordan

- We worked on "Sarah", a project that utilizes deep learning based natural language processing techniques to classify Arabic news.- Due to the lack of relevant datasets concerning Arabic, we developed a data collection pipeline using newspaper3k and django_cron to form a dataset of 10,000 news articles annotated into two classes.- Implemented a data preprocessing pipeline using NLTK and BeautifulSoup to deal with the particularities of the Arabic language.- Experimented with different classification models using Python, Scikit-Learn and Tensorflow.- Deployed and containerized the data pipeline and model as a microservice behind a RESTful API that powered a reporting dashboard using Django. The final model achieved an accuracy of 88%.

Homs, Syria

[Paid/Employment]- Worked on three projects using vanilla web technologies focusing on performance. Got recognized by Awwwards.com, merit can be found in the attachments.[Voluntary]- Organized/co-organized local programming events targeting youth, fresh graduates as well as IT professionals in the community.- Created graphics for social media posts and advertisement.Hours worked per week: 20.

Homs, Syria

Worked on multiple web projects, including: cogis-dgs.com, iso-cert.org, and it-advice.net. (not maintained anymore)Hours worked per week: 10.

Master Of Science - Ms, Computer Science For Autonomous Systems, 4.93/5.0 (Excellent With Honors)

Thesis (distinction): Extending Sparse Dictionary Learning for Adversarial Robustness. Advisor: Dr. habil. András Lörincz & Dr. Dávid.

Bachelor'S Degree, Informatics Engineering, Software Engineering Specialisation, 81.437% (Very Good)

Activities and Societies: Competitive Programming Contest. Thesis: Deep Learning for Cyber Security. Advisor: Suhel Hammoud, PhD. Key.