Accepted for Poster, Presentation & Proceedings at: 3rd International Workshop on AI for Quantum and Quantum for AI (AIQxQIA 2025), ECAI 2025, Bologna, Italy, 25–30 October 2025

Embedding-Aware Quantum-Classical SVMs for Scalable Quantum Machine Learning

🚀 Key Discovery: Vision Transformer (ViT) embeddings unlock quantum machine learning advantage. This is the first systematic evidence that the choice of embeddings determines quantum kernel success, showing a fundamental synergy between transformer attention and quantum feature spaces.

🔗 Project Resources

• Project Page: Embedding-Aware Quantum
• GitHub Repository: QuantumVE
• Research Paper: Embedding-Aware Quantum-Classical SVMs
• Dataset on HuggingFace: QuantumEmbeddings
• Interactive Demo: Colab Notebook

🎯 Breakthrough Results

• Fashion-MNIST: +8.02% accuracy vs classical SVM
• MNIST: +4.42% accuracy boost
• Embedding Insights: ViT embeddings enable quantum advantage; CNN features degrade performance
• Scalability: 16-qubit tensor network simulation via cuTensorNet
• Efficiency: Class-balanced k-means distillation for quantum data preprocessing

Project Architecture

QuantumVE/
├── data_processing/     # Class-balanced k-means distillation procedures
├── embeddings/          # Vision Transformer & CNN embedding extraction
├── qve/                 # Core quantum-classical modules and utilities
└── scripts/             # Experimental pipelines with cross-validation
    ├── classical_baseline.py           # Traditional SVM benchmarks
    ├── cross_validation_baseline.py    # Cross-validation framework
    └── qsvm_cuda_embeddings.py         # Our embedding-aware quantum method

🚀 Quick Start

1. Environment Setup

# Create conda environment
conda create -n QuantumVE python=3.11 -y
conda activate QuantumVE

# Clone and install
git clone https://github.com/sebasmos/QuantumVE.git
cd QuantumVE
pip install -e .

# For Ryzen devices - Install MPI
conda install -c conda-forge mpi4py openmpi

2. Download Pre-computed Embeddings

MNIST Embeddings:

mkdir -p data && \
wget https://huggingface.co/datasets/sebasmos/QuantumEmbeddings/resolve/main/mnist_embeddings.zip && \
unzip mnist_embeddings.zip -d data && \
rm mnist_embeddings.zip

Fashion-MNIST Embeddings:

mkdir -p data && \
wget https://huggingface.co/datasets/sebasmos/QuantumEmbeddings/resolve/main/fashionmnist_embeddings.zip && \
unzip fashionmnist_embeddings.zip -d data && \
rm fashionmnist_embeddings.zip

3. Run Experiments

Single Node:

# Classical baseline with cross-validation
python scripts/classical_baseline.py

# Cross-validation framework  
python scripts/cross_validation_baseline.py

# Our embedding-aware quantum method
python scripts/qsvm_cuda_embeddings.py

Multi-Node with MPI:

# Run with 2 processes
mpirun -np 2 python scripts/qsvm_cuda_embeddings.py
mpirun -np 2 python scripts/cross_validation_baseline.py

🔬 What Makes This Work?

Our key insight: embedding choice is critical for quantum advantage. While CNN features degrade in quantum systems, Vision Transformer embeddings create a unique synergy with quantum feature spaces, enabling measurable performance gains through:

1. Class-balanced distillation reduces quantum overhead while preserving critical patterns
2. ViT attention mechanisms align naturally with quantum superposition states
3. Tensor network simulation scales to practical problem sizes (16+ qubits)

🤝 Contributing

We welcome contributions! Help us advance quantum machine learning:

1. Fork the QuantumVE repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Submit a pull request with detailed description

Areas for contribution:

• New embedding architectures (BERT, CLIP, etc.)
• Additional quantum backends
• Performance optimizations
• Documentation improvements

🙏 Acknowledgements

This work was supported by the Google Cloud Research Credits program under award number GCP19980904.

📄 License

CC BY-NC-SA 4.0

📚 Citation

Paper

@inproceedings{ordonez2025embedding,
  title={Embedding-Aware Quantum-Classical SVMs for Scalable Quantum Machine Learning},
  author={Ord{\'o}{\~n}ez, Sebasti{\'a}n Andr{\'e}s Cajas and Torres, Luis Fernando Torres and Bifulco, Mario and Duran, Carlos Andres and Bosch, Cristian and Carbajo, Ricardo Simon},
  booktitle={3rd International Workshop on AI for Quantum and Quantum for AI (AIQxQIA 2025), ECAI 2025},
  year={2025},
  month={October},
  address={Bologna, Italy},
  note={Accepted for Poster, Presentation \& Proceedings},
  url={https://arxiv.org/abs/2508.00024}
}

---

🌟 Star us on GitHub if this helps your research! 🌟