Collecting and evaluating structured data using classical Machine Learning or Deep Learning algorithms to explore and visualize trends, detect outliers or to make predictions.
Classifying, segmenting and detecting images using self build or pretrained models. Applications can be found for example in Face Detection or Medical Diagnosis.
Using the latest transformer models to analyze and process text. This includes tasks like Machine Translation, Sentiment Analysis and Question Answering to name a few.
Implementing state of the art GANs for unsupervised learning and generative modelling to create new content such as images.
Applying Deep Q-Learning and Policy Gradient Methods to train agents how to successfully interact with their environment. Applications imply Autonomous Driving and Algorithmic Trading.
Deploying models using various cloud platforms or Docker images which can run for example on Kubernetes clusters on premise.
Here is a list of my certificates. Click on the name for more information!
Here is a growing list of sample applications. All of these applications run as docker images on my personal server. Note that none of the applications is trained to perfection as this would require significant hardware resources and computation time. Instead the applications are made to show what can be done. All of the docker images are available on Docker Hub and the source code can be found on GitHub.
This is a model based on Part 3 of the SIC 2018: Skin Lesion Analysis Towards Melanoma Detection challenge aiming to classify skin cancer pictures into 7 classes. This part is also available on Kaggle as Skin Cancer Mnist: HAM 10000 dataset. The model uses a pretrained ResNet152V2 network which has been fine-tuned on a TPU on Google Colab using the dataset above.
This is a collection of six Transformer models built and trained from scratch on a GPU to translate between English, French and German. The application also includes a simple language detector to determine the input language. I used the WMT19 and WMT15 datasets to train the models.