MerLin: A Discovery Engine for Photonic and Hybrid Quantum Machine Learning

TL;DR

The paper addresses the fragmentation and lack of scalable, hardware-aware software for photonic and hybrid quantum machine learning by introducing MerLin, a differentiable, strong-simulation framework tightly integrated with PyTorch/Scikit-learn. It combines a differentiable linear-optical simulator (SLOS) with a modular Quantum Layer, data-encoding options, and a hardware bridge to run parts of models on real photonic hardware, positioning MerLin as a co-design tool. A key contribution is the benchmark-driven reproduction platform that replicates eighteen state-of-the-art QML works, providing reproducible experiments and enabling ablations to disentangle data-, model-, and hardware-driven gains. The framework demonstrates substantial speedups in simulation, preserves learning behavior when porting gate-based models to photonic representations, and offers a practical pathway for systematic, cross-modality QML development and hardware-aware benchmarking. Overall, MerLin serves as a practical discovery engine that links algorithms, benchmarks, and hardware for near-term photonic QML progress.

Abstract

Identifying where quantum models may offer practical benefits in near term quantum machine learning (QML) requires moving beyond isolated algorithmic proposals toward systematic and empirical exploration across models, datasets, and hardware constraints. We introduce MerLin, an open source framework designed as a discovery engine for photonic and hybrid quantum machine learning. MerLin integrates optimized strong simulation of linear optical circuits into standard PyTorch and scikit learn workflows, enabling end to end differentiable training of quantum layers. MerLin is designed around systematic benchmarking and reproducibility. As an initial contribution, we reproduce eighteen state of the art photonic and hybrid QML works spanning kernel methods, reservoir computing, convolutional and recurrent architectures, generative models, and modern training paradigms. These reproductions are released as reusable, modular experiments that can be directly extended and adapted, establishing a shared experimental baseline consistent with empirical benchmarking methodologies widely adopted in modern artificial intelligence. By embedding photonic quantum models within established machine learning ecosystems, MerLin allows practitioners to leverage existing tooling for ablation studies, cross modality comparisons, and hybrid classical quantum workflows. The framework already implements hardware aware features, allowing tests on available quantum hardware while enabling exploration beyond its current capabilities, positioning MerLin as a future proof co design tool linking algorithms, benchmarks, and hardware.

Figures (3)

• Figure 1: MerLin architecture for photonic quantum machine learning. (A) Classical data encoding and photonic circuit configuration define the quantum model. (B)MerLin integrates PyTorch-based optimization with photonic-native execution through a logical-to-photonic bridge, differentiable quantum layers, and hardware-oriented simulation, with optional inference on Quandela photonic QPUs. (C) Measurement strategies and detectors behaviour expose full quantum states or partially measured observables. (D) The resulting amplitudes or probability distributions are returned as classical outputs for downstream machine-learning tasks.
• Figure 3: Quantum Bridge between gate-based circuit and photonic circuit, enabling interoperability between gate-based and photonic frameworks.
• Figure : QuantumLayer initialization and training using MerLin and a PyTorch-native optimization loop