Data

Code conv2d from scratch

Asked at

Microsoft

1Give yourself 5 minutes

2Answer out loud, not in your head

3Then compare with the answer below

Reference answer

Use a four-loop naive implementation (batch, out-channel, y, x -> sum over in_channel, k_y, k_x) and then progress to im2col + GEMM. Discuss stride, padding, dilation, groups, NCHW vs NHWC layout, memory contiguity and how PyTorch's dispatcher selects cuDNN's implicit GEMM. Be ready for the follow-up: "rewrite this in einops notation" or "make it backward pass." Senior interviewers reward: "the actual production path uses cuDNN's implicit GEMM; here's when we beat it (small batch, custom kernel)."

Hiring now

Get the Microsoft interview, not just the prep

landed applies to roles like these for you, on each company’s own site, with answers written for the role.

Get 5x interviews10 free auto-applies. No card.

Or practice this question first

Next question

Bias-variance trade-off
Bagging vs boosting vs Random Forest
Handling imbalanced data
Metric choice for a 0.1%-positive fraud model
Supervised vs unsupervised vs reinforcement learning
Closed-form linear regression vs gradient descent