We want to use the full capabilities of both Qiskit and pyTorch to develop hybrid quantum-classical machine learning algorithms (e.g. meta-learning for quantum circuits with classical neural nets https://arxiv.org/abs/1907.05415 , SchNet with quantum interactions https://arxiv.org/abs/1706.08566 , or emeddings with classical ML as input for quantum circuits). This means that Qiskit circuits should be embedded in pyTorch as a function that can handle backpropagation. While other frameworks already attempt to provide such an interface, they don't allow to define your circuit in native Qiskit language and thus prohibit the use of all it's amazing tools (even gate decomposition is often not supported).
Abstract
We want to use the full capabilities of both Qiskit and pyTorch to develop hybrid quantum-classical machine learning algorithms (e.g. meta-learning for quantum circuits with classical neural nets https://arxiv.org/abs/1907.05415 , SchNet with quantum interactions https://arxiv.org/abs/1706.08566 , or emeddings with classical ML as input for quantum circuits). This means that Qiskit circuits should be embedded in pyTorch as a function that can handle backpropagation. While other frameworks already attempt to provide such an interface, they don't allow to define your circuit in native Qiskit language and thus prohibit the use of all it's amazing tools (even gate decomposition is often not supported).
Members
@kareldumon
email:dumonkarel@gmail.com
@patrickhuembeli
email:patriq@veryquantum.com
Deliverable
A module and a notebooks showcasing it with e.g. meta-learning, SchNet
GitHub repo
https://github.com/BoschSamuel/QizGloria