TutorialsDestiny

Quantum Boltzmann Machines (QBMs)

Introduction

Quantum Boltzmann Machines (QBMs) are quantum extensions of classical Boltzmann Machines, which are energy-based probabilistic models used for unsupervised learning. Unlike classical versions, QBMs leverage quantum superposition and entanglement to represent complex probability distributions more efficiently.

They are particularly powerful for:

  • Sampling complex distributions faster than classical methods
  • Feature extraction for deep learning
  • Optimization tasks where energy-based modeling is crucial

Mathematical Foundation

A classical Boltzmann Machine defines the probability of a state x as:\[P(x)=e−E(x)Z\]

where:

  • \(E(x)\) = Energy of state x
  • \(Z=∑xe−E(x)\) = Partition function (normalization factor)

In a Quantum Boltzmann Machine, the energy function is replaced with a Hamiltonian H:\[ρ=e−βHZ\]

where:

  • \(ρ \)= density matrix (quantum probability distribution)
  • \(β\) = inverse temperature parameter
  • \(H\) = Hamiltonian encoding weights and biases
  • \(Z=Tr(e−βH) \)= quantum partition function

This allows the QBM to model quantum probability distributions instead of classical ones.

Structure of a QBM

  • Visible Units (classical input/output layer): Encodes observed data
  • Hidden Units (quantum layer): Encodes latent features using qubits
  • Hamiltonian (Energy Function): Defines interactions (weights + biases) between visible and hidden units
  • Quantum Sampling: Uses quantum mechanics to sample probability distributions more efficiently

Workflow of Training a QBM

  1. Initialize Hamiltonian parameters (weights + biases)
  2. Encode visible units (data) into qubits
  3. Evolve system under Hamiltonian → obtain quantum state
  4. Sample states using quantum measurement
  5. Update parameters via gradient-based optimization
  6. Repeat until convergence (energy minimized)

Applications of QBMs

✅ Quantum-enhanced generative models – learning distributions from data
✅ Quantum optimization – solving NP-hard problems faster
✅ Quantum-inspired deep learning – acting as hidden layers for QNNs
✅ Drug discovery & materials science – sampling molecular distributions

Visual Diagram

👉 A workflow diagram of QBMs showing:

  • Visible layer (input)
  • Hidden quantum layer (qubits + Hamiltonian)
  • Quantum sampling + parameter updates

➡️ Next: Hybrid Quantum–Classical Algorithms