Add distributed_qml_cc README summary

Author: jsenellart

README.md

@@ -46,6 +46,7 @@ and any per-file or per-directory notices for details.
 | Adversarial Learning. Source: [lu_2020](https://arxiv.org/abs/2001.00030) | We observed on a photonic model, just like the authors of the paper on a gate-based model, that quantum classifiers are vulnerable to direct and transferred adversarial attacks but adversarial training is also effective against specific attack types. MNIST classification (1 vs 9): - Clean accuracy = 98%<br>- Adversarial accuracy (BIM, $\epsilon=0.1$) = 15%<br>- Adversarial accuracy post-adversarial training (BIM) = 95% |
 | [Photonic Quantum Memristor](papers/qrc_memristor/). Source: [selimovic_2025](https://arxiv.org/abs/2504.18694) | We developed a photonic quantum reservoir computing architecture with the addition of a quantum memristor which acts as a feedback loop (memory). We noticed that in both cases the use of the memristor enhances the non-linear capabilities of the model and this leads to improved performance compared to the non-memristor case and some classical benchmarks. The results are similar to those presented in the paper and the error is 5 times smaller on the learning task of the NARMA dataset.|
 | [Limitations of Amplitude Encoding on Quantum Classification](papers/AA_study/). Source: [Wang_2025](https://arxiv.org/abs/2503.01545) | The authors proved and showed numerically the main limitations of amplitude encoding. We observe the same results with a photonic architecture for the simple synthetic datasets and popular image-based datasets. We used the [Photonic Quantum Convolutional Neural Networks with Adaptive State Injection](papers/photonic_QCNN/)'s QCNN architecture for our tests. Our Merlin model seems more stable over the iterations. By just replacing the amplitude encoder by an angle encoder on the simple synthetic datasets, we are able to correctly distinguish both classes. This result show that a user guide for an encoding choice depending on the dataset could be quite useful.|
+| [Distributed QML via Classical Communication](papers/distributed_qml_cc/). Source: [hwang_2024](https://arxiv.org/abs/2408.16327) | Reduced reproduction (3 seeds, 1000 iter, L ∈ {3,5,7,9}) of the 8D synthetic binary classification benchmark with non/NC/CC/QC schemes. The paper's qualitative ordering and the central claim that **CC-DQML closely matches QC-DQML** are reproduced both in a gate-model PyTorch simulator and in a **faithful photonic MerLin translation**. Gate model at L=9: NC 87.6 ± 0.7%, CC 99.2 ± 0.4%, QC 99.8 ± 0.3%. Photonic MerLin (angle encoding, `m=8/n=3` per chip; CC adds a classical-feedforward channel; QC = one doubled `m=16/n=6` chip): NC 88.2 ± 2.9%, CC 95.2 ± 2.5%, QC 98.5 ± 0.3%. An iso-parameter classical MLP baseline (98.5 ± 0.7%) is also included.|
 | [Neural Quantum Embedding: Pushing the Limits of Quantum Supervised Learning](papers/nn_embedding/). Source: [Hur_2024](https://arxiv.org/abs/2311.11412v2) | The authors introduce a novel way to encode classical data on quantum computers using. Indeed, a QML model is separated into two parts: an embedding and classifying circuit. We optimize both of those sections one after the other. The embedding is optimized by training a classical model that takes the classical features as inputs and generates the parameters for the quantum embedding circuit in order to create maximally distant average encoded states. The MerLin implementation is faster and has the same or better performance across all reproduced figures.  |