This project aims to develop a neural network capable of assisting and expediting steering detection for two-qubit states. Specifically, it generates a compact conclusive polytope for each state, which can be used in the optimization problem described in this paper.
The dataset folder contains the Julia scripts used to generate the datasets. Two different encodings are used to create the conclusive polytopes:
- Mother Polytope (
mp): Represents the conclusive polytope as a subpolytope of a larger "mother" polytope. This encoding allows the polytope to be represented as a binary vector. - Free Vectors (
fv): Encodes the conclusive polytope using the angles in polar representation.
You can find my datasets in this google drive folder
The ML_training folder contains the scripts for training the models and tuning hyperparameters. Two different encoding-specific architectures are used:
- Mother Polytope Encoding: This encoding uses a multi-label classification network.
- Free Vectors Encoding: This encoding uses a regression network.
For each encoding, we experiment with two architectures:
- Single Network: Directly predicts the conclusive polytope.
- Branching Network: First predicts if the state is steerable and then generates the conclusive polytope based on this initial prediction.