Coinbase·Data Scientist·Technical Phone Screen·Intermediate

Intermediate

May 2026

Summary

Coinbase data scientist interview had at least one question that was more of a manual computation exercise than anything conceptual. You basically have to do neural network math by hand, layer by layer, which I was not expecting at this level.

Questions Asked (1)

Given a small neural network's weights, biases, and activation functions, manually compute the output for a specific input vector by walking through each layer's matrix multiplication, bias addition, and activation step.

Algorithms & Data StructuresTechnical Trade-offs

Author's notes

I knew the mechanics but doing it by hand under pressure is a different thing.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clearly restating the network architecture and the given input vector. Then, for each layer, compute the pre-activation values (z = Wx + b) and apply the activation function to get the layer output, which becomes the input to the next layer. Finally, present the output of the last layer as the network's prediction.

Pro tip: Double-check your matrix multiplication by verifying dimensions and using a systematic approach (e.g., row-by-column dot products). Also, mention that in practice you'd use a library like NumPy or PyTorch, but manual computation helps debug and understand the model.

1. Understand the network architecture

Identify the number of layers, the size of each layer, the weight matrices, bias vectors, and activation functions for each layer.

2. Compute first layer pre-activation

For the first layer, compute z1 = W1 * x + b1, where x is the input vector, W1 is the weight matrix, and b1 is the bias vector.

3. Apply activation function

Apply the activation function (e.g., ReLU, sigmoid, tanh) element-wise to z1 to obtain the first layer output a1.

4. Repeat for subsequent layers

For each subsequent layer i, compute zi = Wi * a_{i-1} + bi and apply the activation function to get ai, until the final layer.

5. Output final result

The output of the last layer is the network's prediction. Present it clearly, and if applicable, interpret it (e.g., probabilities for classification).

Key Points to Mention

Matrix multiplication and bias addition: z = Wx + b
Activation functions: ReLU, sigmoid, tanh, and their element-wise application
Layer-by-layer forward propagation
Importance of checking dimensions for compatibility
Potential use of vectorization for efficiency
Interpretation of final output (e.g., softmax for multi-class classification)

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