Deep Learning Interview Questions

Deep learning questions ask you to implement the components a framework normally hides: an activation, a normalisation layer, an attention computation, a training step. The point is not to reproduce a library but to show that you understand what the library is doing, particularly around shapes, masking and numerical stability.

What interviewers probe

Shape errors are the most common failure, so narrating the dimensions at each step is both good practice and a visible signal. Interviewers at companies building models probe specifically for the details that matter in production — why attention scores are scaled, why a causal mask is required in a decoder, why normalisation is placed where it is. Being able to connect an implementation detail to its effect on training or inference cost is what marks real experience.