en.Wedoany.com Reported - Swedish startup FirstQFM unveiled a machine learning platform based on a Quantum Foundation Model (QFM) at the ISC High Performance 2026 conference, designed to optimize Quantum Reservoir Computing (QRC) systems. The platform achieved a 56.1% zero-shot sequence-level win rate in financial time series forecasting.
Quantum reservoir computing, as a hybrid sequence modeling framework, utilizes low-depth quantum circuits as high-dimensional feature generators. Unlike traditional implementations that use a single static reservoir, FirstQFM's platform customizes the reservoir by learning contextual information, adapting to the physical state of the underlying processor and the specific characteristics of the prediction problem. Its technology incorporates two core workflows: "problem-aware" and "device-aware." The former analyzes the mathematical structure of the data stream to adjust the reservoir's internal memory and nonlinear curves; the latter monitors the real-time working environment of the quantum processor, adjusting the reservoir based on qubit topology, gate calibration constraints, background crosstalk, and real-time noise vectors.
This Alpha version system was evaluated on 41 daily financial return prediction tasks, covering individual stocks, global indices, crypto assets, and commodities. In zero-shot evaluations, FirstQFM's QRC architecture achieved a lower mean squared error (0.000485 MSE) and higher directional accuracy compared to leading time series foundation models developed by Google, Amazon, and Salesforce. The initial reservoir was generated on the EuroHPC-supported Leonardo supercomputer using the NVIDIA cuQuantum SDK and cuTensorNet library, operating at the edge of classical simulability. To validate performance on larger, non-simulable reservoirs, the team conducted final benchmark tests on Rigetti Computing's multi-chip superconducting quantum hardware, boosting average directional prediction accuracy to 54.74% and achieving a peak single-sequence MSE reduction of 52.95% on major indices such as DAX 30 and Dow 30.
FirstQFM has opened its Beta system to selected pilot partners for processing multivariate enterprise time series. The Beta architecture includes a hardware-aware stabilization layer that dynamically adjusts the feature extraction loop to accommodate changes in qubit physical properties. The enterprise deployment strategy encompasses both cloud and on-premises directions: the on-premises module will leverage NVIDIA NVQLink to establish low-latency connections between local GPU servers and quantum system controllers, enabling enterprise operators to switch between direct prediction and reusable feature layers via natural language control.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
