Solving computational problems using trainable quantum feature maps
Patent Information
- Authority / Receiving Office
- EP · EP
- Patent Type
- Applications
- Current Assignee / Owner
- PASQAL SAS
- Filing Date
- 2024-07-19
- Publication Date
- 2026-05-27
AI Technical Summary
Existing quantum machine learning models face challenges in accurately solving complex computational problems like differential equations due to fixed feature encoding architectures, which limit the representation space and require prior knowledge of optimal encoding configurations.
The introduction of trainable quantum feature maps, where a set of trainable parameters acts directly on the generator Hamiltonian, allows for the dynamic tuning of feature maps during training, optimizing eigenfrequencies and improving the model's ability to approximate solutions to complex problems.
This approach enables quantum models to learn optimal basis functions and spectral properties, leading to more accurate and efficient solutions for computational problems like differential equations, without increasing the depth of quantum circuits.
Smart Images

Figure EP2024070524_23012025_PF_FP_ABST