Guess, SWAP, Repeat : Capturing Quantum Snapshots in Classical Memory

--In this work, we introduce a novel technique that enables the observation of quantum states without directly measuring, and thereby destroying them. Our method enables the observation of multiple quantum states at different points within a single circuit, one at a time, allowing them to be saved into a classical memory without direct measurement or destruction. These states can then be accessed on demand by downstream applications during execution, introducing a dynamic and programmable notion of quantum memory that supports modular, non-destructive quantum workflows. The primary contribution of this work is a hardware-agnostic, machine-learning-driven framework for capturing quantum'snapshot' i.e. non-destructive estimate of quantum state at arbitrary points within a quantum circuit, and enabling their classical storage and later reconstruction, akin to memory operations in classical computing. This capability is critical for real-time introspection, debugging, and memory functionality in quantum systems, yet it remains fundamentally challenging due to the no-cloning theorem and the destructive nature of quantum measurement. This work introduces a'guess-and-check' methodology that utilizes fidelity estimation via the SW AP test to guide state reconstruction. We introduce both gradient-based deep neural networks and gradient-free evolutionary strategies to estimate quantum states using fidelity alone as the learning signal. We implement and validate a key component of our framework on current IBM quantum hardware, achieving high-fidelity ( 1 . In simulation, our models achieve an average fidelity of 0 .999 These reconstructed quantum states can be stored classically and later reloaded into quantum circuits, providing a realistic path toward long-term, non-volatile quantum memory, establishing a practical and generalizable method for quantum state storage, and laying the foundation for future quantum memory architectures. I NTRODUCTION Capturing a quantum snapshot, i.e., the act of identifying a quantum state without destroying it, is currently beyond the reach of practical quantum technology. Likewise, there is no reliable method to implement quantum memory to store an arbitrary quantum state indefinitely. Y et, these capabilities are desirable for the maturation of quantum technologies. They promise transformative potential in areas such as quantum Both authors contributed equally to this research.