quant-phSep 24, 2026

MQSS-Selector: RL-Guided Pass Selection for an MLIR Compilation Pipeline

Authors: Andre Youssefi, Ercüment Kaya, Minh Chung, Jorge Echavarria, Laura B. Schulz, Martin Schulz

Organizations: Leibniz Supercomputing Centre · Technical University of Munich · Munich Quantum Valley · Argonne National Laboratory

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

High Performance Computing (HPC) and Quantum Computing (QC) systems are increasingly converging towards unified High Performance Computing-Quantum Computing (HPCQC) infrastructures, driven by a growing need to bridge classical and quantum workflows, which affects all levels of the system stack, from the hardware to compilers and runtimes, all the way to applications. However, today's QC devices are still in the Noisy Intermediate-Scale Quantum (NISQ) era, are error-prone and resource-limited, and therefore require specialized optimizations and topology mappings to achieve sufficient fidelity. This places special emphasis on proper compilation and optimization within the overall quantum software stack. Many existing stacks remain fragmented, with separate components responsible for device selection, compiler-pass optimization, and job queue scheduling. This paper proposes a unified, learning-based selector that integrates these disparate stages into a cohesive framework. Our proposed selector scheme leverages reinforcement learning and deep learning models that can be extended to simultaneously optimize multiple objectives -- such as fidelity, compilation time, and scheduling latency -- while dynamically adapting to circuit characteristics and device conditions.

Figures & tables

Figure 1 . Illustration of the scheduling pipeline integrating device selection and phase ordering (i.e., compiler pass-sequence optimization) for a quantum program execution.
Figure 2 . a) Conceptual design of the unified scheduler (V1) - Hybrid Supervised + RL, where device selection and phase ordering are handled by interlinked modules. b) Conceptual design of the unified scheduler (V2) - Fully RL-based, combining device selection and phase ordering into a single action space for RLtraining.
Figure 3 . Decay schedules over training epochs. The DAEP trigger probability pDAEPp_{DAEP} , the DAEP magnitude coefficient αDAEP\alpha_{DAEP} and the entropy coefficient αH\alpha_{H} decay to zero, so that guidance and exploration vanish and the actor converges on exploiting the critic’s estimates. The learning rate η\eta decays to a δ>0\delta>0 and thus remains active throughout training.
Figure 4 . Prediction analysis on initial models for the 128 MQT-Bench program set
Figure 5 . Prediction analysis on initial models for the 2,880 RB program set
CategoryMetricGuidance (DAEP)No Guidance
Finish Action AnalysisSamples154154
Avg. total rewards4.320.44
Finish called153/1540/154
Finish ++1460
Finish −-70
Executable programs95.5%42.1%
Table 1 . Chemistry Dataset – Validation Summary comparing DAEP and No Guidance models

Explore similar work

CardsList

  1. Fidelity-Aware Scheduling of Quantum Circuits on Multi-QPU Systems

    Sep 9, 2026Innocenzo Fulginiti, Antonio Tudisco, Salvatore Zammuto +7Quantum CircuitsSchedulers

    Technical University of Munich, Munich, Germany · Department of Electronics and Telecommunications, Politecnico di Torino, Turin, Italy · MQSC, Garching near Munich, Germany +1

  2. CO-MAP: A Reinforcement Learning Approach to the Qubit Allocation Problem

    May 13, 2026Ankit Kulshrestha, Xiaoyuan LiuQubitOffline Reinforcement Learning

    Fujitsu Research of America · Santa Clara, CA - 95054

  3. QAP-Router: Tackling Qubit Routing as Dynamic Quadratic Assignment with Reinforcement Learning

    May 12, 2026Kien X. Nguyen, Ankit Kulshrestha, Ilya Safro +1Quantum CircuitsQubit

    Department of Computer and Information Sciences, University of Delaware, USA · Quantum Lab, Fujitsu Research of America, USA