Resilience: This video demonstrates a capability for identifying when confidence in neural network outputs should be reduced because the input is somehow anomalous (e.g., comes from a different distribution compared to training data, includes a previously unseen class, or presents objects in a very unlikely context).