multi-objective-quantum-workflow

name

multi-objective-quantum-workflow

description

Multi-objective optimization methodology for quantum computing workflows, combining compilation strategy selection, noise suppression, and error-mitigation. Based on QBalance framework (arXiv: 2605.02966) and action-space engineering for RL-based circuit routing. Use when: designing quantum compilation pipelines, optimizing NISQ device execution, selecting error-mitigation strategies, or formulating multi-objective quantum workflow problems.

Multi-Objective Quantum Workflow Optimization

Core Concept

Near-term quantum workloads involve coupled decisions across compilation, noise suppression, and error mitigation. Frame these as finite multi-objective strategy-selection problems over circuits, backends, and transformation policies.

Problem Formulation

minimize  f(c, b, t) = w1·error + w2·latency + w3·cost
subject to t ∈ T, b ∈ B, c ∈ C

Strategy Selection Framework

Step 1: Define Candidate Strategies

Each strategy: (layout_policy, routing_policy, basis_gates, noise_suppression, error_mitigation)

Step 2: Score with Survival-Product Error Proxy

survival_product = ∏_g (1 - ε_g)

Lightweight ranking before expensive circuit execution.

Step 3: Bayesian Candidate Ordering

score(c) = E[w · φ(c)] + β · σ(c)

Feature vector + uncertainty estimate for exploration-exploitation tradeoff.

Step 4: Non-Dominated Selection

Apply Pareto dominance filtering. Select from the Pareto front.

Action-Space Engineering for RL-Based Routing

For RL circuit routing in DQC architectures:

1. State-dependent actions: depend on current qubit placement
2. Action masking: prune invalid actions, reduces space by 10-100x
3. Modular decomposition: separate placement, routing, execution

Distributionally Robust Control

Use Sinkhorn discrepancy for uncertainty sets around noise distributions:

• Combines observed data with prior knowledge
• Convex and tractable for LQ control
• Robust to distributional shifts in quantum gate noise

Practical Workflow

1. Characterization: Profile backend, circuit, estimate baseline error
2. Strategy Search: Generate candidates → score → Bayesian ordering → execute top-K
3. Selection: Build Pareto front → select → execute → update model