Overviews

Goal

Hi! This page is intended for user to understand the overall concept of inspeqtor in the high-level first. You might find this page useful when you reviews the interaction between modules and functions offered by inspeqtor.

We catergorized characterization and calibration of the quantum device into multiple phase. You might not necessary needs to do or understand every phases and chose to work with specific phases.

sequenceDiagram
    participant User
    participant Model as Predictive Model
    participant Device as Quantum Device
    note over User, Device: Characterization
    loop 
    User ->> Device: Perform experiments
    Device ->> User: Data
    User ->> Model: Characterization
    opt Selection strategy
        Model ->> User: Select new experiments
    end
    end
    note over User, Device: Calibration
    loop Optimization
    User <<->> Model: Find the control that <br/> maximize fidelity
    end
    User ->> Device: Deploy calibrated control
    Note over User, Device: Operational
    loop Operating and monitering
    User ->> Device: Use device
    User ->> Device: Check the quality
    end

• Experimental Phase is a preparation of the characterization of the quantum device. It might dictate the constraint of your control calibration too.
• Characterization Phase
• Control Calibration Phase

Note

We would like to remind user that inspeqtor is a framework. We provide some opinions of how to do things in the characterization and calibration task. Thus, inspeqtor provides user with a varities of utility functions which is designed to be easily replaced by custom function from the user. You don't have to use everything, just what you need 😉

Experimental Phase

Diagram below shows the sequence of interaction between, the user, inspeqtor, quantum device, and file system. Please refer to tutorial of how to work with data and experiment using inspeqtor in this tutorial.

sequenceDiagram
    participant User
    participant Control as Control Sequence
    participant Device as Quantum Device<br/>(Real or Simulator)
    participant Data as ExperimentData
    participant Storage as File System

    User->>Control: Define atomic control action
    User->>Control: Create ControlSequence
    Control->>User: Validate & return sequence

    alt Real Hardware
        User->>Device: Setup the device
        Note over Device: Physical quantum device<br/>with real noise & decoherence
    else Simulation
        User->>Device: Setup Hamiltonian & Solver
        Note over Device: Local simulation<br/>with modeled noise
    end

    loop For each sample (e.g., 100x)
        User->>Control: Sample parameters
        Control->>User: Return control params
        User->>Device: Execute with params
        Device->>User: Return expectation values
        User->>Data: Store row with make_row()
    end

    User->>Data: Create ExperimentData
    Data->>Storage: Save to disk
    Storage->>User: Load ExperimentData back

    Note over User,Storage: Same data format regardless<br/>of real device or simulator

Characterization Phase

Now let us proceed with the characterization with the experimetal data at hand. The sequence diagram below hilight the overview of the interactions between entities. Basically, user will have to load the data from the file system into a memory and perform necessary operations to prepare a data. Then, user can use them to perform characterization using the selected model. Please check the tutorial for how we can use inspeqtor for these tasks.

sequenceDiagram
    participant User
    participant Data
    participant Optimizer
    participant Models

    Note over User,Data: Data Preparation
    User->>Data: Prepare & Split Data
    Data-->>User: Return Training & Testing Data

    Note over User,Models: Model Initialization
    alt Statistical Model (DNN)
        User->>Optimizer: Define Model & Loss Function
        Optimizer->>Models: Initialize Parameters
    else Probabilistic Model (BNN)
        User->>Optimizer: Define Model, Prior, Guide, & SVI
        Optimizer->>Models: Initialize SVI State
    end

    Note over Optimizer,Models: Model Training
    User->>Optimizer: Start Training Loop

    loop For each epoch
        Optimizer->>Models: Update parameters
        Note right of Optimizer: Validates against testing data
    end

    Optimizer-->>User: Return Trained Model

Alternative Characterization phase

sequenceDiagram
    participant User
    participant Strategy as Abstract Strategy
    participant Model
    participant Device

    User->>Strategy: Prepare Experiment

    loop Characterization Loop
        Strategy->>Strategy: Select next experiment parameters
        Note right of Strategy: This can be Random (Open-Loop) or <br> Model-Informed/Adaptive (Closed-Loop).

        Strategy-->>User: Recommend experiment

        User->>Device: Perform experiment
        Device-->>User: Measurement data

        User->>Model: Update/Characterize Model

        Model-->>Strategy: Provide Posterior Model (if adaptive)

        Strategy->>Strategy: Check termination condition
    end

    Strategy-->>User: Return Final Characterized Model

Control Calibration Phase

Together with the characterization phase, we bundle the tutorial, please check!.

sequenceDiagram
    participant User
    participant Optimizer
    participant CostFunction as Cost Function <br/> (e.g., Avg. Gate Infidelity)
    participant Model as PredictiveModel
    participant Device as Quantum Device<br/>(Real or Simulator)

    Note over User, Model: Starts with Trained Predictive Model from Characterization

    User->>CostFunction: Define(Target Gate, PredictiveModel)
    User->>Optimizer: Start Optimization(CostFunction, Initial Params)

    loop Optimization Steps
        Optimizer->>CostFunction: Evaluate(current_params)
        CostFunction->>Model: Predict(current_params)
        Model-->>CostFunction: Return Expectation Values
        CostFunction-->>Optimizer: Return Loss (Infidelity)
        Optimizer->>Optimizer: Update Parameters
    end

    Optimizer-->>User: Return Optimized Control Parameters

    %% alt Benchmarking
    User->>Device: Execute(Optimized Params)
    Device-->>User: Return Measured Fidelity
    User->>Model: Predict(Optimized Params)
    Model-->>User: Return Predicted Fidelity

Alternately, we can use closed-loop optmization (no need for the gradient) with Bayesian optimization. As of the current implementation, we demonstrate a minimal usage of Bayesian optimization inside the workflow of inspeqtor in tutorial