Quantum machine learning (QML) holds great promise, but practical implementations often encounter a major bottleneck before quantum computation even begins. Classical feature vectors must first be prepared and encoded into quantum states. As data dimensionality and the number of qubits increase, this step can dominate end-to-end execution time and limit the scalability of QML workflows.

In this session, we will introduce Quantum Data Plane (QDP), a GPU-accelerated data plane being developed in Apache Mahout to make quantum data encoding more efficient and scalable. QDP treats data encoding as a dedicated execution layer rather than a one-off preprocessing task. We will explain its motivation, overall architecture, integration with Apache Mahout, and how GPU parallelism can help address the encoding bottleneck.

Drawing on our implementation experience, we will share key design choices and lessons learned from building QDP. We will discuss where GPU acceleration provides the greatest value, the limitations and trade-offs of this approach, and how the project may evolve through open-source collaboration.

Attendees will leave with a practical understanding of data preparation challenges in QML and a broader perspective on how Apache Mahout can connect classical machine learning infrastructure with emerging quantum computing workflows.

Speakers:

Jie-Kai Chang: Apache Mahout PMC Member

Jie-Kai Chang is actively involved in open-source communities and project governance, with a focus on high-performance computing, GPU acceleration, and machine learning infrastructure. He also participates in the Ray and KubeRay communities, exploring distributed computing, cloud-native AI, and AI workload management on Kubernetes. His current work addresses data-encoding bottlenecks in quantum machine learning and advances the design and implementation of Apache Mahout’s Quantum Data Plane (QDP).

Guan-Hua Wen: Apache Mahout Committer

Research and Development Intern at Microsoft