Adrian Hoffmann Email & Phone Number

Hi, I grew up on a farm and managed to train a deep RL agent without human data to beat friends and family at Jass (a popular Swiss card game) - Yeah, I can imagine you didn't expect to read that combination. Besides knowing how to milk a cow and playing Jass, I love to apply myself to problems that excite me and where I can help people (definitely including research). This is especially true for ideas in interpretable AI as well as software and AI that aid people in their daily (work) life. Feel free to reach out if you have any ideas. Here are some projects my interests have taken me to:- With Bau-Apps GmbH we noticed that there is a huge potential to make life easier for construction managers with organizing issues at their construction sites. Since then, we are closely working together with them to aid them with mobile and desktop solutions with the ultimate goal of saving them time. Check out our tool at baupen.ch.- Another place where I could help with digitising is bookkeeping on farms (in my case somewhat obviously) – specifically my parents’ farm. Combining a desktop application with a mobile Android app I digitised the selling of eggs. This includes tracking orders, deliveries, and automatic billing, all with special attention to ease of use.- I am very thankful to the people at the Secure, Reliable, and Intelligent-Systems Lab headed by Prof. Martin Vechev at ETH Zurich for the chance to implement large parts of their initial release of their ETH Robustness Analyzer for Neural Networks (ERAN).- In my Master's Thesis I set out to improve preprocessing of EEG data with downstream gaze prediction in mind. For that, I trained a deep model (VAE) on completely raw EEG data to isolate components that could then be used to predict a person's gaze from their brain activity alone. Labelled data then furthermore allowed to gain insight what information different components seem to encode.