1st round as AI engineer (robotic company): Explain why log loss works for Logistic regression? What will be the loss be for multiclass? What loss will you use for multi-class labeling? What is data imbalance? What will you do with imbalanced data? In Sklearn they give one parameter for unbalanced data where they use internally? A major difference between CNN and ANN? How does CNN work? What is pooling? How does it work? What is regularization? What is overfitting? What is dropout? How will it work at training and inference time? What are precision and recall? Some questions from a case study.
