Debugging AI
summary
The talk explains why debugging and verification for AI systems differ from traditional software and why testing machine learning is still immature. It gives a brief overview of AI and machine learning, describes characteristics of AI solutions, and presents several approaches for verifying machine learning systems to ensure models learn the right behavior.
AI-written summary · about
chapters (13)
- 0:00Start
- 15:54Playing Chess, playing Go
- 17:59Another typical problem: Computer vision
- 18:23Recap: Properties of AI problems
- 18:46Problems with unknown truth
- 20:22ML replicates bias in the data
- 23:12AI systems break fundamental patterns we developed as an industry
- 27:23Model requirements
- 28:56Use BI, data visualization, statistics, Excel
- 31:09Make sure you know the data
- 32:35Work with test sets
- 34:08These techniques are not generally applicable
- 38:46Add a safety net
related talks
43mDebugging AI
45mTesting Applications That Use AI
45mHow to Test Artificial Intelligence and Machine Learning Solutions?
29mTesting and Assuring AI-Based Systems
32mQuality engineering in the AI age: From testing AI to leveraging GenAI
41mAI and cyber, or why your vacuum cleaner (still)...
01h 04mWhat Testers Need from Machine Learning and Artificial Intelligence
01h 05mNavigating the Security and Privacy Landscape of Modern AI