2019 – TCAV Method for Explainable Artificial Intelligence (XAI)

Tine Kolenik introduced the field of explainable artificial intelligence (XAI), what machine bias is and why it is important to tackle it. He presented the efforts of the Google Brain team to produce methods for understanding how black box models make decisions and focused on one such method, Testing with Concept Activation Vectors (TCAV). TCAV describes the process of getting quantitative evaluation of how significant human concepts are for neural network models for making certain predictions.