Pulse-based quantum evolution kernel learning for molecular property prediction using graph-based embeddings

Accurate prediction of fundamental chemical properties such as bond dissociation energies and reaction barrier heights remains a central challenge in chemistry and materials science. Classical machine learning methods have achieved notable success, yet they often face scalability limitations and difficulty capturing complex, high-order correlations in molecular graphs. Quantum Machine Learning (QML) offers a fundamentally different approach by embedding molecular systems into high-dimension representational spaces. Most existing QML models rely on parameterized quantum circuits optimized variationally, which can suffer from trainability challenges and noise sensitivity on current Noisy Intermediate-Scale Quantum (NISQ) hardware.
Alternatively, kernel-based approaches within QML embed classical data into quantum states and evaluate similarity measures between data points, analogous to classical kernel methods. While such quantum kernels can be implemented using parameterized circuits, they can also be constructed within a Hamiltonian dynamics framework, where classical graph-structured inputs are encoded into many-body interactions and similarity is derived from quantum evolution and measurement statistics on the evolved state. Because the underlying Hilbert space grows exponentially with system size, these evolution-based kernels can generate similarity measures that are difficult to reproduce classically in full generality while remaining naturally compatible with NISQ platforms.
Pasqal develops programmable neutral-atom quantum processors capable of analog time evolution and pulse-level control. Building on this capability, Pasqal has demonstrated Quantum Evolution Kernels (QEK) for graph classification tasks. This project extends QEK to chemistry-driven regression problems, specifically molecular property prediction, while exploring physically motivated pulse parameterization to enhance kernel expressivity, hardware compatibility and execution time. The anticipated outcome is a validated quantum kernel regression pipeline aligned with neutral-atom hardware, positioning Pasqal competitively in scientific machine learning applications and contributing to advances in catalyst design, reaction optimization, and materials discovery.

Faculty Supervisor:

Viki Kumar Prasad;Victor Drouin-Touchette

Student:

Partner:

PASQAL Canada

Discipline:

Physics

Sector:

Information and cultural industries

University:

University of Calgary

Program: