APPLIED ARTIFICIAL
INTELLIGENCE LAB
VALUES OF THE GROUP

Scientific Integrity Before Performance

• Correct methodology over impressive results
• Honest reporting of negative or inconclusive results
• Clear explanation of limitations
• If results look too good to be true, we investigate and not celebrate
• Do not avoid discussion about failure cases
• Do not fish seeds

Reproducibility Is a Must

• Reproducible experiments are mandatory
• Transparent code, data handling and evaluation
• Github repository is mandatory for journal paper
• Clear documentation of methodology, assumptions and parameters
• Every result should be reproducible by others
• No manual figure or table editing without scripts
• It worked on my PC is a red flag

Data Efficiency

• Simple baselines before complex models
• Thoroughly consider the data
• Be careful with augmentations
• Datasets for VPD are challenging, keep that in mind

Critical Thinking and Individual Independence

• Asking why and how is essential
• Challenge the methodology and assumptions
• Try to propose alternatives, do not just execute instructions
• Our goal is understanding, not immediate agreement
• Unquestioned decisions are a red flag

Constructive and Respectful Collaboration

• Open discussion and respectful disagreement
• We help each other to succeed
• Some bad words are not meant as they might sound

Clear Communication and Shared Responsibility

• Clear expectations and explicit decisions
• Speak up early about problems or uncertainty
• If needed, speak with supervisor in private
• Silence does not equal agreement
• Beware of silent delays

Sustainable Work and Long Term Improvement

• Sustainable pace over constant urgency
• Learning and skill development need time
• Mental and physical well being matters, be honest and speak up
• Deadlines should be respected
• Burnout is not a badge of honor

Application Oriented Responsibility

• Be aware of real world complexity and consequences
• Be responsible with results interpretation
• Avoid overclaiming, for example clinical deployment soon

Good Practices

• Learn from mistakes, do not be afraid to make them
• Reflect on what works and what does not, try to understand why
• Update methods, tools and processes
• Code review is our friend
• Stagnation is a symptom of failure

Fair Recognition and Transparency

• Decide about authorship early
• Contribute with strong skills, improve weaknesses in challenging tasks
• Do not have implicit expectations about credit