← Coinbase Interview Insights

Coinbase·Machine Learning Engineer·Technical Phone Screen·Senior

Senior
Apr 2026

Summary

Coinbase ML engineer interview with a neural network forward-pass computation question, the kind of thing that sounds straightforward until you're actually grinding through sigmoid math by hand under pressure.

Questions Asked (1)

Q1

Given a small fully connected neural network with specific weights and biases, compute the forward pass step by step using sigmoid activation. Show the pre-activation values (z = Wx + b) and post-activation values (a = sigmoid(z)) for each layer.

Algorithms & Data StructuresTechnical Trade-offs
Author's notes

The math itself isn't hard but doing it by hand while explaining your reasoning out loud is a different thing.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clearly stating the dimensions of each weight matrix and bias vector, then compute the pre-activation z for each layer using matrix multiplication and addition. Apply the sigmoid function element-wise to get the activations, and present the final output with a brief interpretation.

Pro tip: Double-check your matrix multiplication order and sigmoid calculations; a small arithmetic error can cascade. Also, mention that sigmoid outputs are probabilities, which is useful for binary classification tasks common in fraud detection at Coinbase.

1. Understand the network architecture

Identify the number of layers, neurons per layer, and the input features. Note the weight matrices and bias vectors for each layer.

2. Compute pre-activation for the first layer

Calculate z1 = W1 * x + b1, where x is the input vector. Show the matrix multiplication and addition step by step.

3. Apply sigmoid activation for the first layer

Compute a1 = sigmoid(z1) element-wise. Show the sigmoid formula and calculate each value.

4. Repeat for subsequent layers

For each hidden layer, compute z = W * a_prev + b, then a = sigmoid(z). Continue until the output layer.

5. Present final output and interpret

State the final output activations and explain what they represent (e.g., class probabilities). Optionally, discuss computational efficiency.

Key Points to Mention

  • Matrix dimensions and compatibility for multiplication
  • Sigmoid function formula: σ(z) = 1 / (1 + e^{-z})
  • Element-wise application of activation function
  • Importance of bias terms in shifting the activation
  • Numerical stability considerations (e.g., overflow in exp)
  • Interpretation of output as probabilities for binary classification

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.