IBM has released the beta version of its Qiskit Transpiler Service, available to all IBM Quantum Premium Plan users. The service uses artificial intelligence (AI) and IBM cloud resources to speed up transpilation tasks in quantum computing. Transpilation is a process where an abstract quantum circuit is rewritten to match the constraints of a specific quantum device. The new service allows users to experiment with AI-powered transpiler passes, which may be faster and produce better results than traditional methods. The authors of the article are David Kremer, Juan Cruz Benito, Ismael Faro, and Robert Davis.
Introduction to Qiskit Transpiler Service
IBM Quantum has recently announced the beta release of the Qiskit Transpiler Service, now available to all IBM Quantum Premium Plan users. The Qiskit Transpiler Service is a Python library that utilizes the resources of the IBM Cloud to provide users with the latest transpilation capabilities from the Qiskit SDK. Transpilation is a crucial process in quantum computing, where an abstract quantum circuit is rewritten to match the constraints and characteristics of a specific quantum device. This service allows users to experiment with new and improved AI-powered transpiler passes, which are tools that may be faster and produce better results than traditional transpilation methods.
AI-Powered Transpiler Passes
The AI-powered transpiler passes were first introduced at the IBM Quantum Summit 2023. These passes are used for circuit routing and circuit synthesis, which are foundational tasks in quantum circuit transpilation. AI methods have been shown to produce shorter and shallower circuits than the standard heuristic methods currently available in Qiskit. Furthermore, AI-powered transpilation can be much faster than Boolean satisfiability (SAT) solvers and other optimization methods. These capabilities were initially made available to Premium Plan users as part of the experimental Qiskit Transpiler Service alpha release. With the beta release, these capabilities have been further improved.
Getting Started with the Qiskit Transpiler Service
To start using the Qiskit Transpiler Service, users can integrate the service into their current Qiskit patterns and workflows. The service can be used with or without AI. For instance, if a user wants to keep their experiment consistent with the heuristic methods they’ve used so far, they can set the AI parameter to False. This will give them the standard Qiskit SDK transpiler pipelines via the cloud. On the other hand, if a user wants to try the AI-powered passes, they can set the AI parameter to True. This will give them transpilation that is built upon the standard Qiskit SDK transpiler pipelines, but integrates the AI-powered passes for improved results.
AI Routing and Circuit Synthesis Passes
The AI-powered transpiler passes can also be used as building blocks in transpiling routines. They essentially work as drop-in replacements for traditional Qiskit passes in certain transpiling tasks, such as circuit routing and circuit synthesis. By combining AI-powered passes with heuristic passes, users of the Qiskit Transpiler Service can reduce two-qubit gate counts by an average of 42 percent. The AI routing pass acts both as a layout stage, in which the system determines how qubits on your circuit map to the qubits on the device, and as a routing stage, where the system inserts SWAP gates to make the two-qubit gates in the circuit compatible with the device’s coupling map.
AI-Powered Transpiler Passes: Under the Hood
The AI-powered transpiler passes are enabled by reinforcement learning (RL) methods. RL is a machine learning paradigm in which algorithms learn some optimal “policy” by interacting with an environment. The model’s goal is to maximize the cumulative reward by finding the right action to take each time it interacts with the environment. RL models are a particularly good fit for transpilation tasks because they are trained by trial and error against a given problem. This means they have the potential to generate algorithms that outperform existing ones. For instance, if we want to train a reinforcement learning algorithm for Clifford circuit synthesis, we start by providing the algorithm with a target Clifford as an input. The algorithm selects gates step by step until the target Clifford is reached. During training time, the algorithm may be penalized for every CX gate, penalized for every 1-qubit gate, and rewarded if the target is reached. This reward function teaches the algorithm to choose the gates that provide the shortest path to the target circuit.
External Link: Click Here For More
