Scientific Workflow for QML

Intermediate Quantum ML Methodology

Created by Pavel · 11.03.2026 at 14:32 UTC

Treat a QML project like an experiment: fix the task, metric, and classical baseline before touching circuits. Then pin shots, seeds, and splits so variance is interpretable rather than mysterious.

The failure mode is retrofitting a story after a lucky run—without uncertainty and resource accounting, neither you nor a reviewer can tell signal from noise. References: [1], [2].

Sources

University approvals: 0

Tasks

Question 1

Which step should come first in a disciplined QML experiment?

Hint

Scientific framing before implementation.

Tune random circuit depth until accuracy looks good

Define problem, metrics, and baseline

Skip baseline and report quantum result only

Pick hardware brand first and derive the problem later

Question 2

If your objective is expectation-based, you usually prefer which primitive?

Hint

Observable average primitive.

estimator

sampler

transpiler

drawer

Question 3

Why is reporting variance/uncertainty important in shot-based experiments?

Hint

Sampling noise.

Because outcomes are stochastic and finite-shot estimates fluctuate

Because variance guarantees quantum advantage

Because uncertainty replaces accuracy

Because uncertainty means implementation is wrong

Question 4

Name one anti-pattern discouraged in Lectures 1-3 style methodology.

Hint

Think over-claiming.

claiming quantum advantage without a baseline

increasing shots

deeper circuits

random restarts

Question 5

A high-quality QML report should ideally include:

Hint

Full experimental context.

Only final accuracy

Task definition, baseline, resource assumptions, and uncertainty

Only circuit image

Only narrative motivation

Question 6

A model has mean accuracy 0.81 with sample standard deviation 0.04 over 16 runs. What is the standard error of the mean?

Hint

$SE = s / \sqrt{n}$.

0.01

0.11

0.02

Question 7

A result shows +1.5% accuracy but uncertainty interval overlaps zero improvement. Best conclusion?

Hint

Distinguish effect size from statistical confidence.

Definitive improvement is proven

Evidence is suggestive but insufficient for a strong improvement claim

The experiment is invalid regardless of protocol

Classical baseline should be removed

Question 8

Which reporting practice most improves reproducibility of QML experiments?

Hint

Reproducibility requires protocol transparency.

Report only best-case metric

Provide seeds, shot budgets, splits, and resource details

Hide hyperparameters to avoid bias

Use different metrics per model

Question 9

Implement standard_error_mean(sample_std: float, n_runs: int) -> float as $s/\sqrt{n}$ for n_runs >= 1. Raise ValueError if n_runs < 1 or sample_std < 0.

Hint

math.sqrt for the denominator.

import math


def standard_error_mean(sample_std: float, n_runs: int) -> float:
    # TODO: validate, return sample_std / sqrt(n_runs)
    pass

Starter code is prefilled; replace TODO blocks with your solution.

Runtime output (stdout/stderr)

2 test cases will be used for grading

Run checks runtime behavior only. Final correctness is evaluated when you submit.

Card Info

Topic: Quantum ML Methodology
Difficulty: Intermediate
Completed: 0 users

Creator

Pavel