Each round typically includes one or two LeetCode-style algorithm questions, usually focusing on data structures like arrays, hash maps, or binary trees.

In addition to coding, they also ask some basic but fundamental machine learning and deep learning questions. For example, they might ask about:

• Explaining how dropout works
• The role of batch normalization
• Differences between SGD and Adam
• How overfitting can be handled

Sometimes, they might dive deeper into topics like activation functions, loss functions, or model regularization techniques. The difficulty is usually moderate, but they do expect clear and structured explanations.

Introduce dropout.

Why Batch Normalization (BN) works. Why BN needs a shift and rescale.