Priyam Srivastava

Research

Quantum sensing and quantum-network control

I use variational optimization to design quantum sensing circuits and classical reinforcement learning to control entanglement generation. Numerical simulation connects these methods to limited control and finite memory coherence.

variational quantum sensing

Variational probe and readout design

I model dipolar-interacting spin systems in PennyLane and JAX and use CMA-ES to optimize probe-preparation and readout circuits. For structured single-parameter phase estimation, I compare sequential and joint optimization using the same finite-depth encoder and shared two-angle product readout. Quantum and classical Fisher information distinguish the information in the probe from the information extracted by the measurement.

For joint magnetic-field and gradient estimation on dipolar spin chains, I optimize the determinant of the classical Fisher-information matrix. In this model, a GHZ probe has a rank-one quantum Fisher-information matrix and cannot identify both parameters independently. At five spins and three circuit layers, variational performance reaches 92% of the best-found quantum Fisher-information determinant benchmark and 4.2 times the standard-quantum-limit determinant.

arXiv:2507.22043 arXiv:2605.03906 QCE 2025

Two training protocols

Sequential: select the probe using a fixed Ramsey readout, then freeze the probe and optimize the readout.

Joint: optimize probe-preparation and readout parameters together.

The deployed encoder and readout resources are matched. The comparison tests how including the readout during probe selection affects sensing performance. These are numerical studies; the optimized settings have not been demonstrated here on a physical device.

Illustration: GHZ sensitivity under dephasing

At fixed interrogation time, independent dephasing suppresses GHZ coherence more strongly as the number of qubits increases. This idealized comparison shows when that loss outweighs the noiseless sensitivity advantage. It is separate from the variational sensing results above.

8
0.00

product uncertainty

0.354

GHZ uncertainty

0.125

product / GHZ uncertainty

2.83x

product GHZ number of qubits

Model: each qubit retains coherence e−d, with dimensionless d = Γt. The locally optimal phase-uncertainty bounds are ed/√N for a product probe and eNd/N for a GHZ probe, normalized to one repetition. Interrogation time and repetition count are fixed; preparation and readout are ideal. A ratio above 1 favors GHZ, and below 1 favors the product probe. This does not compare protocols after optimizing interrogation time or include error correction. Lines connect integer qubit counts. These are model predictions, not experimental data or results from my papers.

reinforcement learning for network control

Link-layer learning and entanglement-swapping protocols

I built a two-layer simulator in PyTorch and Gymnasium. Classical REINFORCE policies control the elementary links, optimizing secret-key rates and supplying entangled Werner pairs to link buffers. I then hold those policies fixed and compare two prescribed network-layer protocols: sequential entanglement swapping and simultaneous SWAP-ASAP. The network-layer protocol is the comparison variable, rather than a learned choice between the two protocols.

SLURM sweeps on Pitt CRC vary external memory coherence over four orders of magnitude. In the fixed-size chain studied, sequential swapping is strongly suppressed at short coherence times and approaches simultaneous-swapping performance at long coherence times. The ratio of external memory coherence time to per-link heralding latency organizes these regimes.

arXiv:2605.04047

Two-layer quantum-network simulation: fixed link-layer learning policies supply entangled pairs, and sequential and simultaneous entanglement-swapping protocols are compared.
Fixed link-layer policies supply entangled pairs to buffers. Separate simulations compare sequential and simultaneous swapping, isolating the effect of network-layer protocol and memory coherence.

AI for quantum networking

I authored the AI for quantum networking section of When AI Meets Quantum Information: A Comprehensive Review, submitted to Progress in Quantum Electronics. I am a contributing coauthor of the full review.

Earlier projects