AI Writings
A collection of hands-on articles researching technology, opportunities and limits of human-centered AI.
Verifying Saliency Maps with a biased Vision Model
Saliency maps are an explanative method for vision models. In my experiment though, there were high inconsistent results.
Explanation techniques for a Random Forest that predicts stroke
Implementing and comparing state-of-the-art explanative methods for tabular data.
Privacy and security in AI/ML
Privacy concerns occur over the whole AI/ML life cycle. This article explores strategies and technology to mitigate risk. Includes a benchmark on homomorphic encryption.
How the network architecture impacts robustness and why this does not matter.
Benchmarking several architectures revealed significant differences in robustness. The story contains a twist: this knowledge won't be that helpful.
The case of harmful explainable AI
Europe wants Trustworthy AI. But what happens if users trust an AI-system? My user study shows the downsides of trust. This study was conducted in light of my master thesis.
Developing a conversational AI with Rasa
Conversational AI enables use cases far beyond the chatbot. Rasa is an engine developers can use to build and deploy their own solution. I explored the tech and share my experiences.