AI Evolves Quantum Circuits, Bypassing Design Limits for More Powerful Computers
Summarize this article with:
Designing effective quantum circuits presents a significant hurdle in the development of scalable quantum computation, as circuit structure profoundly impacts performance and feasibility. Devroop Kar, Daniel Krutz, and Travis Desell, all from the Rochester Institute of Technology, address this challenge with a novel evolutionary approach called the Evolutionary eXploration of Augmenting Circuits (EXAQC). Their research introduces a method that simultaneously optimises gate types, qubit connectivity, parameterisation, and circuit depth, whilst accommodating hardware limitations and noise. This work demonstrates the potential of evolutionary search as a crucial tool for advancing quantum machine learning, offering a systematic route towards circuits that are scalable, tailored to specific problems, and efficient for implementation on available hardware. Initial findings reveal that EXAQC-evolved circuits achieve over 90% accuracy on benchmark classification tasks with modest computational resources, and successfully replicate target circuit states with considerable fidelity. This breakthrough addresses a central challenge in quantum computing, where circuit structure profoundly impacts expressivity, trainability, and feasibility for implementation on quantum hardware. The research introduces the Evolutionary eXploration of Augmenting Quantum Circuits, or EXAQC, a method that simultaneously searches for optimal gate types, qubit connectivity, parameterization, and circuit depth while adhering to hardware and noise constraints. Unlike current methods relying on manual design or fixed heuristics, EXAQC leverages principles from neuroevolution and genetic programming to create circuits specifically tailored to the problem at hand. The work highlights a significant advancement in automated quantum circuit design, moving beyond limitations in scalability, flexibility, and adaptability often found in existing approaches. EXAQC employs an evolutionary algorithm to explore a vast design space, enabling the discovery of circuits that are both problem-aware and hardware-efficient. This innovative framework supports both Qiskit and Pennylane libraries, providing users with comprehensive control over every aspect of the circuit design process. By jointly optimizing circuit structure and parameters, the method overcomes challenges associated with barren plateaus and weak gradient signals, common issues hindering the training of variational quantum circuits. Preliminary results demonstrate the efficacy of EXAQC, with evolved circuits achieving high fidelity scores when emulating target quantum states. The system successfully classifies data from datasets such as Iris, Wine, Seeds, and Breast Cancer, demonstrating its potential for real-world applications in quantum machine learning.
This research establishes evolutionary search as a critical tool for advancing variational quantum algorithms, offering a principled pathway toward scalable and robust quantum circuit design. The ability to discover expressive circuit topologies organically, rather than relying on predefined layers, represents a key innovation in the field. Evolutionary optimisation of quantum circuit genomes and parameters offers a promising avenue for quantum algorithm design A 72-qubit superconducting processor forms the foundation of the Evolutionary eXploration of Augmenting Quantum Circuits (EXAQC) methodology, an evolutionary approach to automated design and training of parameterized quantum circuits. The research directly optimises both circuit structure and parameters for supervised learning tasks and emulation of target circuit behaviour. Quantum circuits are represented as mutable genomes comprising parameterized and non-parameterized quantum gates, allowing for the evolution of circuit depth, gate ordering, qubit connectivity, and entanglement patterns. Circuit parameters undergo optimisation utilising gradient-based learning techniques, while structural modifications are explored through the application of evolutionary operators. This results in a hybrid evolutionary-variational training process that efficiently navigates the complex design space. The method jointly searches over gate types, qubit connectivity, parameterisation, and circuit depth, simultaneously addressing multiple critical design aspects. Importantly, EXAQC respects both hardware limitations and the presence of noise during the evolutionary process, ensuring the generated circuits are practically implementable. The framework supports integration with both Qiskit and Pennylane libraries, affording users complete configuration control over every aspect of the circuit design process. This backend-agnostic design enhances the versatility and applicability of the method across diverse quantum computing platforms. Preliminary results demonstrate that circuits evolved on classification tasks achieve over 90% accuracy on benchmark datasets, even with limited computational resources. Furthermore, these evolved circuits successfully emulate target circuit quantum states with high fidelity scores, validating the effectiveness of the approach. This work establishes evolutionary search as a valuable tool for advancing quantum machine learning and variational quantum algorithms. Evolved quantum circuits demonstrate high accuracy on classification benchmarks, surpassing classical machine learning models Over 90% accuracy was achieved on most benchmark datasets using circuits evolved on classification tasks with a limited computational budget. The research details an evolutionary approach, termed EXAQC, for automated design and training of parameterized quantum circuits. This method simultaneously searches over gate types, qubit connectivity, parameterization, and circuit depth while adhering to hardware and noise constraints. Circuits are represented as mutable genomes comprising parameterized and non-parameterized quantum gates, facilitating the evolution of circuit depth and gate ordering. The framework supports both Qiskit and Pennylane libraries, allowing comprehensive user configuration of every aspect of the circuit design process. Classical features are embedded into quantum states via angle-based encodings, and predictions are derived from designated readout qubits using marginal probability distributions. Through evolution, increasingly entangled input and output registers were observed, leading to improved classification performance across datasets including Iris, Wine, Seeds, and Breast Cancer. Expressive circuit structures emerged organically from the evolutionary process, rather than being pre-defined through entangling layers. The study employed measurement-driven loss functions, such as cross-entropy over marginal readout probabilities, aligning with classical classification objectives. This work highlights evolutionary search as a critical tool for advancing quantum machine learning and variational quantum algorithms, providing a pathway toward scalable, problem-aware, and hardware-efficient quantum circuit design. The research demonstrates the potential for automated discovery of nontrivial circuit topologies capable of achieving high accuracy with limited resources. Neuroevolutionary design of high fidelity quantum circuits is a promising research direction The Evolutionary eXploration of Augmenting Circuits (EXAQC) offers a novel automated approach to designing and training parameterized quantum circuits. This method concurrently optimises gate types, qubit connectivity, parameterisation, and circuit depth, all while adhering to specified hardware and noise limitations. Supporting both Qiskit and Pennylane libraries, EXAQC provides extensive user configuration options and leverages techniques from neuroevolution and genetic programming to achieve efficient circuit design. Preliminary results indicate that circuits evolved using EXAQC can achieve over 90% accuracy on several benchmark classification datasets with reasonable computational demands. Furthermore, the evolved circuits demonstrate a capacity to replicate target circuit states with substantial fidelity. The framework’s adaptability extends beyond any fixed gate set, accommodating nearly all gates supported by standard quantum computing libraries. This work establishes evolutionary search as a valuable technique for advancing quantum machine learning and provides a systematic route towards scalable, problem-specific, and hardware-compatible circuit construction. Acknowledging current limitations, the authors note that EXAQC presently employs a single population and a single objective function for optimisation. Future research will focus on incorporating multiple populations and employing diverse speciation strategies to improve optimisation performance. Additionally, the framework will be extended to accommodate multi-objective optimisation, leveraging the variety of loss metrics already available within EXAQC. Expanding the application of EXAQC to areas such as reinforcement learning, time series forecasting, and complex classification tasks like computer vision will further demonstrate its versatility as a general quantum circuit optimisation framework. 👉 More information 🗞 Investigating Quantum Circuit Designs Using Neuro-Evolution 🧠 ArXiv: https://arxiv.org/abs/2602.03840 Tags: