#exaqc

The Evolutionary Exploration of Augmenting Circuits EXAQC

Rochester Institute of Technology (RIT) quantum information scientists have developed a revolutionary automated quantum circuit design method. By integrating neuroevolutionary and genetic programming to “evolve” quantum systems, Evolutionary eXploration of Augmenting Circuits (EXAQC) overcomes the disadvantages of human-engineered architectures.

This invention by Devroop Kar, Daniel Krutz, and Travis Desell aims to tackle the major impediment to scalable quantum computation: designing circuit topologies that are both high-performing and practical for existing hardware. As the industry moves toward Noisy Intermediate-Scale Quantum (NISQ) technology, the EXAQC paradigm offers a logical, problem-aware path to dependable quantum machine learning.

Complexity of Quantum Design Space

Quantum circuits are tricky and sometimes use “ansatz” layers or manual heuristics. A circuit's expressivity, trainability, and viability are affected by its structure, including depth, gates, and qubit connection.

The “barren plateaus” phenomenon is a major obstacle to training variational quantum circuits. Gradient signals grow so feeble that optimization is nearly impossible, stopping learning. Researchers must also contend with quantum noise and hardware limits, which can quickly impair computation accuracy.

The EXAQC architecture was designed to address these concerns without templates. The technology lets expressive circuit topologies evolve spontaneously through evolutionary search instead of relying on human intuition to choose the best circuit.

EXAQC's “Mutable Genome”

EXAQC's main innovation is depicting quantum circuits as customizable genomes. Parameterized and non-parameterized quantum gates make form the circuit's "DNA" (genomes). Using evolutionary operators, the framework can change circuit structure elements like:

Gate order, circuit depth.

Qubit entanglement and connection.

Gate kinds and parameters.

Thus, training is evolutionary and variational. Gradient-based learning adjusts circuit parameters as the evolutionary algorithm searches the enormous design space for the optimal structural configurations. This dual optimization method ensures expressive and hardware-implementable circuits.

Proven Global Benchmark Performance

Comprehensive testing has proven this evolutionary method works. Supervised learning was done using the 72-qubit superconducting processor-based EXAQC framework. For classical data, angle-based encodings embed features into quantum states. We then employ marginal probability distributions to predict from selected readout qubits, which matches typical classification goals.

The results are great. EXAQC-evolved circuits achieved over 90% accuracy on benchmark classification tasks like the Iris, Wine, Seeds, and Breast Cancer datasets with low computational power, according to preliminary results.

Beyond categorization, the framework simulates target circuit quantum states with significant adaptability. The framework's flexibility for quantum research is confirmed by the circuits' realistic simulation of complex states. Researchers found that input and output registers become more entangled during evolution, which was linked to improved performance across datasets.

Scalable, backend-agnostic solution

The RIT team made EXAQC backend-agnostic to maximize scientific utility. The framework supports industry-standard libraries like Qiskit and Pennylane and allows flexible setup. The circuits can be modified to accommodate practically any set of gates compatible with typical quantum computing platforms due to their versatility.

EXAQC optimizes structure and parameters for scalable, hardware-efficient design. This is crucial because quantum computers develop bigger and more sophisticated, making manual design harder.

The Future: Multi-Objective Evolution

Despite the current success, the researchers believe there is room for improvement. EXAQC uses one population and objective function for optimization. Future studies of many populations and different speciation processes should increase optimization efficiency even more.

The team plans to add multi-objective optimization to the framework. Researchers could utilize the framework's loss metrics to balance different parameters, such as accuracy and circuit depth or noise sensitivity.

Many applications are conceivable using EXAQC. Next, the RIT team wants to use the technology in more complex domains like:

Learn by reinforcement.

Time series forecasting.

Visual computing

To conclude

Kar, Krutz, and Desell's paper shows how evolutionary search helps design variational quantum algorithms. Automating the finding of nontrivial circuit topologies with EXAQC leads to circuits that are well-suited to their problems.