CrackML by @ml.with.umang
Interview questions / ML Fundamentals
ML Fundamentals interview question

How would you train a model from first principles?

How would you train a model from first principles?

mediumconceptEvidence 40/1001 source reportGoogle DeepMind

The 60-second answer

Forward propagation computes activations and loss; backprop applies the chain rule from loss back to each parameter. Cache the local quantities needed for gradients and propagate vector-Jacobian products in reverse order.

Build the answer in this order

1
Give the core idea

Forward propagation computes activations and loss; backprop applies the chain rule from loss back to each parameter.

2
Explain how it works

Cache the local quantities needed for gradients and propagate vector-Jacobian products in reverse order.

3
Compare alternatives

Verify tensor shapes and use gradient checking on small examples when implementing from scratch.

4
State failure modes + validation

The computational cost is typically on the same order as the forward pass.

A useful interview mental model

This is the shape of a strong answer—not a script to memorize.

01Definition
02Mechanism
03Trade-offs
04Failure modes
05When to use

Senior-level signal

  • Discuss numerical stability, mixed precision, and gradient accumulation for large models.
  • Explain how autodiff builds and traverses the computation graph rather than treating backprop as magic.

What the interviewer is really testing

Mechanistic understanding, assumptions, trade-offs, and whether you can turn a definition into a model decision.

Likely follow-up questions

What assumption makes this approach work?
When would you choose the strongest alternative instead?
What production or data failure mode changes your answer?

Common weak-answer patterns

  • Reciting a definition without mechanism or assumptions.
  • Claiming one technique is always better without a data regime.
  • Stopping before failure modes, validation, or deployment implications.