Artificial intelligence
Bring quantum circuits into model learning.
For classification tasks, encode sample features as quantum states and build models using similarity measurements or parameterized circuits. Feed measurement results into classical learning and parameter updates.
Compare similarities, then learn to classify.
Map samples to quantum states, compare their similarities through measurement, and pass these relationships to a classical classifier. Quantum circuits provide feature representation and similarity computation.
Task and data
Define features and labels, then split the data into training, validation and independent test sets.
Quantum feature mapping
Apply the same encoding to two samples and compare their relationship in quantum feature space.
Similarity readout
Estimate pairwise similarities to form a kernel matrix for a classical classifier.
Classical classification
Train a classifier on the kernel matrix to learn decision boundaries, then predict classes for new samples.
Return to feature mapping
Compare encodings and circuit structures using validation performance, then select the model settings for the next iteration.
Bring trainable circuits into the learning process.
Encode features in a quantum circuit and adjust the quantum state with trainable parameters. Measurements produce predictions; a classical optimizer computes the loss and updates parameters across successive training iterations.
Task and data
Define features and labels, then split the data into training, validation and independent test sets.
Encoding and trainable circuits
Feature mapping handles the input; the parameterized circuit adjusts the model representation.
Measurement and prediction
Run the circuit repeatedly to estimate observables and obtain predictions for task evaluation.
Loss and parameter updates
Compare predictions with labels; a classical optimizer computes the loss and updates circuit parameters θ.
Feed into the next circuit iteration
Run the circuit again with the updated θ, continuously recording loss changes and sampling results.
Turn learning tasks into
traceable experiments.
Fix inputs and evaluation methods, recording circuit configurations, training and outputs so that the next adjustment has a clear basis.
From model circuits to hardware execution
Superconducting quantum computing systems use microwaves to prepare and control quantum states, with measurements acquired through the readout chain.
- Data and tasks
- Define features, labels and target metrics, and fix the training, validation and test splits so different models begin with the same inputs.
- Circuits and parameters
- Keep records of encoding, circuit structure and parameter updates, together with execution counts and sampling settings, to track each training iteration.
- Results and feedback
- Compare prediction performance, training stability and total computational overhead against classical models, using the results to guide the next model adjustment.