Learn from quantum data
Predict physical phases and observables from measured quantum states without giving the learner hidden Hamiltonian parameters or exact statevectors.
Quantum AI · evidence first
Learn physical properties without hiding uncertainty.
EnviTrace is developing quantum machine-learning workflows that treat measurements, uncertainty, drift, and abstention as parts of one scientific decision.
Evidence flow
Parameterized quantum states
Finite preparations · incompatible observables · physical phases
Single copy
Joint copies
Fidelity kernel
Prediction with a decision boundary
Predict · quantify · monitor drift · abstain
Why quantum data
Test the use case where quantum learning has a natural role.
When the source data are quantum states, useful properties may depend on measurements that cannot all be collected from the same copy.
That makes acquisition strategy part of the learning problem: which observables to measure, whether joint measurements are worthwhile, and when the evidence supports a prediction.
Quantum structure creates potential for useful learning methods. It does not, by itself, establish quantum advantage.
Budget scarce measurements
Compare single-copy, joint two-copy, and quantum-kernel acquisition strategies using the same counted state-preparation ledger.
Keep uncertainty actionable
Calibrate prediction sets, monitor distribution drift, request more evidence, and abstain near physical phase boundaries.
Demonstration problem
Many-body property learning under finite measurements.
The current Q-Sure study constructs four-site XY-chain states, classifies three interior phases and identifies a boundary class, predicts incompatible signed observables, calibrates uncertainty, and abstains near learned boundaries.
A finite-shot quantum fidelity kernel learns above the four-class prior, but it does not beat the strongest measured-feature control and does not pass the full reliability gate.
Review the Q-Sure evidenceFrozen simulator evidence
112 states
30 streams
Measured-feature control
0.898
macro-F1
post-hoc strongest control
Q-Sure-selected policy
0.870
macro-F1
passed
Finite-shot quantum kernel
0.817
macro-F1
1 / 30
Evidence ladder
Make maturity visible.
Available now
Offline evidence
Reproducible simulators, fixed seeds, held-out evaluation, negative controls, preparation accounting, and explicit claim gates.
Not yet run
Hardware evidence
Timestamped calibration, queue and compilation costs, repeated executions, and matched classical controls remain required.
Not established
Advantage evidence
A credible claim must survive end-to-end resource accounting and beat the strongest relevant measured-feature and classical baselines.
Explore EnviCloud
Explore quantum assurance in action.
Use the public Q-Sure demo to inspect how evidence, measurement budgets, and abstention shape a quantum-ML decision.
- Public demo
Q-Sure Quantum Assurance
Compare measurement strategies, uncertainty, resource budgets, rerouting, and abstention in an evidence-aware quantum workflow.
Open demo (opens in a new tab) - Public demo
Information Theory Explorer
See how patterns and useful detail emerge in one- and two-dimensional data.
Open demo (opens in a new tab)
Build evidence together
Bring a scientific property-learning problem.
We are interested in problems where quantum data, measurement cost, uncertainty, and scientific validity must be evaluated together.
Discuss a quantum AI workflow