The second was a coding interview. My question was: Given a 2D trajectory, use a series of (x, y, t) tuples to represent the position at different times, where t monotonically increases, and the trajectory between two points is a straight line. 1) The first question was relatively simple: given any t, find the coordinates (x, y). 2) After completing the first question, he followed up by asking: given any point on a plane, find the shortest path between it and this trajectory. If the trajectory has many segments, he asked how to reduce the computation cost. The third was a research interview. I also talked to a research scientist, but this time it was much more serious. He spent about half the time asking questions based on my paper, covering everything from motivation and high-level ideas to specific implementations—basically, anything that came to mind. The other half consisted of some basic machine learning questions. I recall: 1) Explain the principles of dropout / how it's used during training or testing / why it reduces overfit, and provide explanations from different perspectives. 2) What are overfit / underfit, and how are they determined? 3) I can't remember exactly, but it might have involved ensemble.