LLMs Discovering State-of-the-Art Quantum Circuits
This paper introduces an evolutionary framework where Large Language Models (LLMs) are used to design novel quantum circuits. An independent evaluator then verifies and scores these designs, leading to significant improvements in quantum algorithm efficiency across various domains.
Reading in Plain English mode — toggle below to switch
This paper introduces an evolutionary framework where Large Language Models (LLMs) are used to design novel quantum circuits. An independent evaluator then verifies and scores these designs, leading to significant improvements in quantum algorithm efficiency across various domains.
Imagine an evolutionary algorithm where the 'mutation' step is performed by an LLM generating new quantum circuit ideas, and the 'selection' step is an independent quantum simulator evaluating their performance. Over generations, the LLM 'learns' to propose increasingly optimal circuits, much like natural selection refines species.
How LLMs Design Quantum Circuits
- 11. LLM Proposal: A large language model is prompted to generate a quantum circuit design for a specific computational task, adhering to quantum mechanical constraints like reversibility.
- 22. Independent Evaluation: A separate quantum circuit simulator or verifier executes the proposed circuit and calculates its performance metrics, such as Toffoli gate count, qubit count, or computational depth.
- 33. Feedback and Iteration: The evaluation results are fed back into the system, potentially guiding the LLM to refine its proposals in subsequent iterations, mimicking an evolutionary search process to find more efficient designs.
C_optimized = LLM(Cᵢnitial, E(C_proposed))
$C_{optimized}$ is the optimized quantum circuit, $LLM$ is the large language model, $C_{initial}$ is the starting circuit or problem description, and $E(C_{proposed})$ is the evaluation function providing feedback on proposed circuits.
LLMs can be effectively integrated into evolutionary frameworks to autonomously discover novel and highly optimized quantum circuit designs, surpassing human-designed state-of-the-art in several metrics.