A Model of Understanding in Deep Learning Systems

1. 1Identify the target system and its properties to be understood
2. 2Develop an adequate internal model that tracks real regularities
3. 3Establish stable bridge principles to couple the internal model to the target system
4. 4Evaluate the reliability of predictions made by the internal model

ML researchers and AI engineers benefit from this model as it provides a framework for evaluating the understanding of deep learning systems, enabling them to design more effective and reliable models

💡 Deep learning systems can achieve understanding through internal models and bridge principles, but often fall short of ideal understanding

💡 New model of understanding in deep learning: internal models + bridge principles = reliable prediction