Quantum Lab Session 6 Complete Simulation Q18
πŸ€– Before we begin β€” when quantum meets AI
Machine learning has transformed the world: image recognition, language translation, medical diagnosis. But classical neural networks are hitting fundamental limits β€” the models are getting so large that training them requires enormous amounts of energy and data.

Quantum Machine Learning (QML) asks: what if we used quantum circuits instead of classical neurons? A parameterised quantum circuit (PQC) can represent some mathematical functions that would require an exponentially larger classical network. And quantum computers naturally explore high-dimensional spaces β€” which is exactly what machine learning requires.

But here's what makes this simulation different: we'll be honest about the hype. QML is genuinely promising for specific problems β€” but claims that "quantum AI will replace classical AI" are premature. Understanding what QML can and can't do is exactly what separates a real quantum scientist from a headline reader.
πŸŒ€ The honest state of QML in 2025
QML has shown quantum advantage on carefully constructed toy problems. Real-world advantage on practical ML tasks has not been demonstrated. The field is growing rapidly β€” but so is the skepticism about near-term feasibility. The most honest assessment: QML will likely be most useful for ML tasks where the data itself comes from quantum systems (molecular simulation, materials design), not for tasks like image classification where classical ML already excels.
πŸ€– Quantum ML Β· Session 6 Β· Q18

Quantum Machine Learning

Parameterised quantum circuits as neural networks β€” watch a quantum model learn to classify data, compare its convergence to a classical network, and get an honest assessment of where QML helps today vs where it's still theoretical.

🧠 Classical vs Quantum
πŸ”„ PQC Architecture
πŸ“ˆ Train a QNN
βš–οΈ Honest Assessment
πŸ† Badge
🧠

Classical neural nets

Layers of weighted connections. Each neuron computes a weighted sum + nonlinearity. Billions of parameters. Train via backpropagation on GPUs.

⚑

Quantum neural nets

Parameterised quantum circuits replace classical layers. Rotation angles are the "weights". Measured expectation values give predictions. Trained via parameter shift rule.

πŸ“Š

Potential advantage

Quantum circuits can represent some function families more compactly. Natural fit for quantum data (chemistry, physics). Kernel methods may benefit.

⚠️

Honest limits

No demonstrated advantage on real-world classical ML tasks. Barren plateau problem limits trainability. Classical ML is extremely competitive. Hype is real.

πŸ€–
Wizzy Β· Quantum Guide
Classical neural networks are made of layers of neurons with adjustable weights. A quantum neural network replaces these with parameterised quantum gates β€” rotation gates with adjustable angles. Both learn by adjusting their parameters to minimise a loss function. The key question: do the quantum versions offer any advantage?
πŸŒ€ The core analogy between classical and quantum
Classical: input β†’ weighted sum β†’ activation function β†’ output. Quantum: input encoded as qubit state β†’ rotation gates (angles = weights) β†’ measurement β†’ output. The "activation function" in QML is measurement collapse. Training updates rotation angles using the parameter shift rule (quantum equivalent of backpropagation).

Classical Neural Net vs Quantum Neural Net

🧠 Classical Neural Network
Input β†’ weighted neurons β†’ output
⚑ Quantum Neural Network
State encoding β†’ PQC β†’ measurement
Classical NN
Parameters: weights & biases
Training: backpropagation
Hardware: CPU/GPU
Scalable: yes (millions of params)
Data: classical
Quantum NN (PQC)
Parameters: rotation angles
Training: parameter shift rule
Hardware: quantum processor
Scalable: limited (NISQ era)
Data: classical or quantum
πŸ€–
Wizzy Β· Quantum Guide
A Parameterised Quantum Circuit (PQC) has three parts: data encoding (classical input β†’ qubit angles), variational layers (trainable rotation gates β€” these are the "weights"), and measurement (collapse to classical output). Press Next Layer to walk through the circuit structure.
πŸŒ€ Why quantum circuits can be expressive
A single rotation gate on n qubits can entangle all n qubits simultaneously β€” creating correlations that would require exponentially many classical parameters to represent. This "expressibility" is what gives PQCs their theoretical advantage. However, in practice the barren plateau problem makes these circuits very hard to train as they get deeper.

Parameterised Quantum Circuit β€” Layer by Layer

Press to walk through the PQC structure β€” from classical input encoding to measurement output.
πŸ€–
Wizzy Β· Quantum Guide
Now watch a quantum neural network learn to classify data. The training plot shows both the quantum model (teal) and a classical network (orange) converging on the same classification problem. Watch the data points get correctly classified as training progresses. Which converges faster?

Training a QNN β€” Quantum vs Classical

Classification accuracy
⚑ Quantum (QNN)
50%
🧠 Classical (NN)
50%
Generate a dataset then train both models. Watch how quickly each converges.
πŸ€–
Wizzy Β· Quantum Guide
This is the most important phase: an honest look at where QML actually helps today. The field has a hype problem β€” many claims exceed evidence. But there are genuine areas where quantum advantage is plausible. Understanding the difference is what real quantum scientists do.
πŸŒ€ The barren plateau problem β€” why QML is hard to train
As a PQC gets deeper, the gradients of its parameters vanish exponentially β€” they become so tiny that optimisation becomes impossible. This "barren plateau" problem is one of the most serious challenges in QML. It means that naive QML doesn't scale β€” deeper networks become untrainable. Active research on circuit structure, initialization, and local cost functions aims to address this.

Honest Assessment β€” Where QML Helps vs Hype

βœ… QML likely helps:
Quantum data (molecular simulation, materials) β€” quantum computers speak the same language as the data
Quantum kernel methods β€” well-defined mathematical advantages for specific data distributions
Small-data regimes β€” PQCs may generalise better with limited training data on some problems
Hybrid pipelines β€” quantum subroutines embedded in classical ML workflows
❌ QML doesn't help (yet):
Image classification, NLP β€” classical deep learning is mature and extremely efficient; no demonstrated QML advantage
Large datasets β€” encoding classical data into quantum states is slow and introduces overhead that erases any benefit
Deep QNNs β€” barren plateau problem makes training impossible beyond shallow circuits
General speedup claims β€” "quantum AI will be exponentially faster" is not supported by current evidence
Research maturity β€” where QML stands in 2025
Quantum kernels
Promising
Quantum data ML
Strong
Classical task QML
Weak
Generative QML
Early
The bottom line: QML is an exciting research frontier with genuine theoretical promise. But honest science requires acknowledging that practical advantage on real-world classical ML tasks has not been demonstrated. The strongest near-term case for QML is in quantum data β€” processing outputs from quantum sensors, quantum simulations, and quantum communication systems.
πŸ€–
Wizzy Β· Quantum Guide
🎊 Session 6 β€” Quantum Applications β€” Complete! You've covered the three most commercially important near-term quantum applications: molecular simulation for drug discovery (Q16), combinatorial optimisation (Q17), and machine learning (Q18). You're now equipped to separate quantum hype from quantum reality!
🧠 What you learned in Session 6
  • Quantum drug discovery (Q16): VQE finds molecular ground states on NISQ hardware. Classical computers require 2α΄Ί memory for N electrons; quantum computers store this naturally in N qubits. Priority: Alzheimer's, cancer, nitrogen fixation.
  • QAOA optimisation (Q17): Alternating cost and mixer operators use quantum interference to find good solutions to NP-hard problems. Hybrid classical-quantum. Applications: Indian Railways, logistics, power grid, healthcare.
  • Quantum ML (Q18): Parameterised quantum circuits are quantum neural networks. Trained via parameter shift rule. Genuinely promising for quantum data β€” honest assessment: no demonstrated advantage on classical ML tasks yet.
  • The barren plateau problem limits QNN depth β€” gradients vanish exponentially in deep circuits. Active research area.
  • All three Session 6 algorithms are hybrid quantum-classical β€” quantum hardware evaluates the hard part; classical computers optimise. This is the NISQ-era strategy.
πŸ€–

Quantum ML Pioneer! Session 6 Complete!

You understand QML β€” and you can tell the difference between quantum hype and quantum reality!

πŸ€– WhizzStep Quantum Lab Β· Session 6
This certifies that
Student Name
has mastered Quantum Applications β€” Drug Discovery, QAOA Optimisation & Quantum ML
QML Pioneer
PQC Expert
Session 6 βœ“
πŸ“– Quantum Vocabulary
PQC KEY

Parameterised Quantum Circuit. A quantum circuit with adjustable rotation angles β€” the quantum equivalent of a neural network with trainable weights.

QNN NEW

Quantum Neural Network. A machine learning model implemented as a PQC. Input is encoded as qubit states; output is measurement expectation values.

Like a classical neural net, but weights are rotation angles and neurons are qubits.
Parameter shift rule NEW

The quantum equivalent of backpropagation. Computes gradients of rotation angles by running the circuit twice with Β±Ο€/2 shift β€” exact gradients on real hardware.

Barren plateau KEY

As a PQC gets deeper, gradients vanish exponentially β€” making training impossible. One of the most serious challenges in QML. Active research area.

Quantum kernel

A kernel function computed using quantum circuits. Maps classical data to quantum feature space. Theoretically advantageous for some data distributions; practically limited by data encoding cost.

Data encoding

Converting classical input data into qubit states. The bottleneck of QML β€” encoding N classical data points requires O(N) quantum operations, which can erase quantum advantage.

Key Concepts from Q18

PQC

⚑ Quantum weights

Rotation angles in a PQC play the same role as weights in a classical NN. They're optimised to minimise a loss function using the parameter shift rule β€” exact gradients on quantum hardware.

Barren plateau

πŸ“‰ The trainability wall

Deep PQCs have exponentially vanishing gradients. This is the fundamental scalability challenge in QML β€” random initialisation leads to flat loss landscapes that optimisers cannot navigate.

Honest take

βš–οΈ Hype vs reality

QML for quantum data: strong theoretical basis. QML for classical tasks (images, text): no demonstrated advantage. Real quantum scientists distinguish between promising research and breathless press releases.

Session 6

πŸŽ“ Applications complete

You now understand all three leading quantum application areas: chemistry (Q16), optimisation (Q17), ML (Q18). Sessions 7-8 cover India's quantum race, quantum vs classical comparisons, and the grand finale.