| name | canonical-quantization-neurons |
| category | quantum-computing |
| description | Canonical quantization methodology for constructing quantum neuron models from classical Hamiltonians — a principled framework for quantum machine learning primitives |
| trigger_words | canonical quantization, quantum neurons, quantum machine learning, activation observable, Hamiltonian simulation, power of one qumode, Schroedingerization, quantum function approximation |
| arxiv_id | 2607.05000 |
| date | 2026-07-07 |
Canonical Quantization of Neurons
Paper
Title: Canonical quantization of neurons
arXiv: 2607.05000
Date: 2026-07-06
Category: quant-ph, cs.LG
Core Methodology
Applies canonical quantization — a systematic procedure for constructing quantum models from classical Hamiltonians — to the fundamental ML primitive: the neuron.
Key Innovation
- Neuron as composition: Views a neuron as composition of an energy function and an activation function
- Quantization step: Replaces energy function with quantum Hamiltonian and applies activation function through matrix functional calculus
- Activation observable: Results in an activation observable that can be measured on input quantum state
- Function approximation: Learns unknown observable from labeled quantum data
Training Algorithms
Hybrid quantum-classical algorithms for:
- Measuring activation observable using: power of one qumode, Schroedingerization
- Gradient estimation using: classical random sampling, Hadamard test, Hamiltonian simulation
- Squared loss error estimation via quantum algorithms
Results
- Quantized neurons exhibit enhanced expressive capabilities vs classical neurons
- Principled framework for constructing quantum ML primitives
- Foundation for quantum neural architectures
Implementation Pattern
1. Define classical neuron: σ(E(x)) where E is energy, σ is activation
2. Quantize energy: E → Ĥ (quantum Hamiltonian)
3. Apply activation via matrix functional calculus: σ(Ĥ)
4. Measure activation observable on input quantum state |ψ⟩
5. Train using hybrid quantum-classical optimization
Reusable Patterns
- Canonical quantization as design principle: Systematic procedure for quantum ML primitives
- Matrix functional calculus for activation: Applying nonlinear functions to quantum operators
- Hybrid training primitives: Hadamard test, power of one qumode, Schroedingerization
- Observable-based learning: Learning unknown observables from labeled quantum data
