STIR: Machine Learning Moderately Controlled Quantum Dynamics
Abstract
Approaches to creating and controlling large-scale quantum systems have primarily focused on two different paradigms. There are highly granular strategies where each subsystem is controlled perfectly and where larger quantum states to be built up from individual and precisely engineered interactions. In such cases O(N) control complexity, i.e. O(N) lines and read-out channels, are needed for an N-body quantum system. Alternatively, there are strategies based on very few pump fields that drive a many-body system into entangled states that have been proposed and leveraged with remarkable success in AMO physics. In these approaches, very few fields (roughly O(1)) are needed for highly nontrivial and nonequilibrium N-body dynamics to emerge. Our goal is to discover a new paradigm that falls outside of the fully controlled, and fully uncontrolled strategies described above. Within both paradigms, physicists and quantum engineers often translate deep ideas and concepts into useful dynamics (e.g. topological surface codes, many-body localization, classical error correction, and chaos). Can we take the physicist out of the equation? Can machine learning allow us to find useful quantum dynamics in an autonomous manner? Can quantum phenomena be engineered with only a moderate level, e.g. O(N1/2), scaling of control complexity? We address these questions be using power tools being developed in the fields of artificial intelligence and machine learning. Designing quantum dynamics with machine learning is challenging due to the complexity of large quantum states. To facilitate a machine learning approach, we will leverage computational techniques from quantum information and condensed matter theory. Tensor and neural network states have shown good performance in calculating the ground state of some many-body quantum system. The success of these approaches comes from the fact that both tensor network and neural network states are powerful and efficient in describing the entanglement in certain many-body systems. As a concrete appilication of current technological importance, we will focus on developing ways that quantum information can be preserved in a many body system despite the presence of error processes. Our effort will lead to an approach that may outperform the current methods for error correction. Furthermore, we will try to understand theoretically the properties of the codes we discover and try to decode their inner working. Our ultimate goal is to implement this platform for experiments, taking the actual qubit configuration as well as the error channels as input, and automatically finding the optimized QEC schemes to protect quantum information in-situ.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 06, 2019
- Source ID
- W911NF1910422
Entities
People
- Amir H. Safavi-Naeini
Organizations
- Army Contracting Command
- Stanford University
- United States Army