Differentiable Quantum Computing for Large-scale Linear Control

Authors: Connor Clayton, Jiaqi Leng, Gengzhi Yang, Yi-Ling Qiao, Ming Lin, Xiaodi Wu

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5.3 Numerical Experiments We conduct a numerical experiment to showcase the correctness of our quantum policy gradient algorithm.
Researcher Affiliation Academia 1Joint Center for Quantum Information and Computer Science, University of Maryland 2Department of Computer Science, University of Maryland 3Department of Mathematics, University of Maryland 4Center for Machine Learning, University of Maryland 5Department of Mathematics and Simons Institute for the Theory of Computing, UC Berkeley
Pseudocode Yes Algorithm 1 Quantum policy gradient Inputs: A, B, Q, R (problem data), K0 SK (initial guess), σ > 0 (step size/learning rate), θ (robustness parameter), N (number of iterations) Output: an approximate solution KN
Open Source Code Yes The code for both methods can be seen at https://github.com/Yiling_Qiao/diff_lqr.
Open Datasets No The paper describes constructing problem instances from physical systems ('a mass-spring-damper system', 'aircraft control problem') and defines the matrices A, B, Q, R directly, but does not provide access information (link, DOI, citation) to a publicly available or open dataset.
Dataset Splits No The paper discusses a 'training process' for the policy gradient method, but it does not specify any exact dataset split percentages, sample counts, predefined splits, or cross-validation setup for training, validation, or testing.
Hardware Specification Yes Both methods run on a classical simulator with Intel i910980XE CPU.
Software Dependencies No The paper mentions that code is available on GitHub but does not explicitly list specific software dependencies with version numbers (e.g., 'Python 3.8', 'PyTorch 1.9') within the paper's text.
Experiment Setup Yes Following a similar setup as in [45], a mass-spring-damper system with g = 4 masses is used for constructing our LQR problem. The state x = [p , v ] R2g contains positions and velocities, with dynamic and input matrices, A = 0 I T T , Q = I + 100e1e 1 , R = I + 4e2e 2 , where 0, I are g g zero and identity matrices, ei is the ith unit vector, and matrix T has 2 on the main diagonal and -1 on the first superand sub-diagonal. ... We set A = [[0, 1], [0, 0.5]] , B = [0, 1] , Q = [[10, 0], [0, 1]] , and R = [0.1].