Shivanshu Siyanwal

Conference paper

Hybrid variational quantum circuits based temperature forecasting with low-latency FPGA inference

N. Khan, S. Siyanwal, K. Islur, A. Tiwari, H. Gupta

IEEE TQCEBT 2026 · published

Overview

A hybrid quantum machine learning approach for 24-day temperature forecasting from daily NCMRWF reanalysis data, using a 7-qubit variational quantum circuit with Strongly Entangling Layers and 50 trainable parameters, trained on a CPU-based quantum simulator, with low-latency inference on an AMD Alveo U55C FPGA kernel.

Problem / motivation

24-day-ahead temperature forecasting is hard: atmospheric dynamics are nonlinear and labelled data is limited. The question was whether a compact variational quantum circuit, trained on daily NCMRWF reanalysis data, could learn useful long-range patterns, and whether its inference could run fast enough for edge use.

Methods

The model is a 7-qubit variational quantum circuit with 50 trainable parameters, using angle encoding to map meteorological variables onto the qubits and two StronglyEntanglingLayers as the ansatz. It was trained, validated and tested in PennyLane on a CPU-based quantum simulator, reaching a test RMSE of 3.01 K and R² of 0.801.

For low-latency edge inference, the frozen model runs on a custom Vitis HLS kernel on an AMD Alveo U55C FPGA; the kernel exchanges batches of input features, pre-trained weights and expectation values with the host over HBM.

Results / outcomes

A hybrid VQC forecaster with low-latency FPGA inference for edge use. Published at IEEE TQCEBT 2026.

My role

Co-author (with first author N. Khan), IEEE TQCEBT 2026.

ProblemApproachAccelerationResult
Schematic: structure of the work (from the write-up's sections)

Project write-up with figures →