Hybrid Quantum AI models for Accelerating Multiphysic Simulations Part 1 | Aditya Seshaditya

Value of the Information & Strength of the Argument

The talk provides significant value by clearly identifying the limitations of classical PINNs (spectral bias, point‑wise evaluation, and loss imbalance) and proposing a concrete hybrid quantum approach to address them. The argumentation is thorough: the speaker justifies each design choice (shallow circuits, two‑layer PQC, Fourier feature mapping) with the goal of understanding the quantum influence on learning. Preliminary results are presented with appropriate caution, and the open questions (quantum contribution, stopping criteria) are explicitly stated. The presentation effectively conveys the potential of quantum‑enhanced physics‑based AI, but the lack of rigorous benchmarking against classical baselines and absence of independent validation temper the claims.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigour is moderate: the work is in an early stage and results are not yet published, but the methodology is grounded in established techniques (PINNs, parameterized quantum circuits, Fourier features) and references are provided via the GitHub repositories and the Quasi digital site. The title is accurate and reflects the content without overpromising. The talk does not cite external literature in detail, but relies on its own implementation and standard simulation tools. The absence of peer‑reviewed sources and the limited dataset size (18,000 points) lower the overall reliability. Adéquation titre/contenu is excellent; the title accurately describes the hybrid quantum AI focus for multiphysics simulations. No comments were provided to analyse.

Title accurately reflects the content: hybrid quantum AI models for multiphysics simulations.

7/10

The talk presents preliminary research with honest acknowledgment of limitations, but lacks peer-reviewed validation and is based on early-stage results.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction to the two key domains: silicon crystal growth and liquid hydrogen cryogenics, and the need for digital twins.
• 10:00Discussion of classical physics-informed neural networks (PINNs), their limitations (point‑wise evaluation, spectral bias, loss imbalance).
• 18:00Presentation of the hybrid architecture: classical backbone with a 16‑qubit, 2‑layer parameterized quantum circuit using Fourier feature mapping.
• 25:00Details of the coupled Navier–Stokes equations and the training procedure using PennyLane, Qiskit AER, and IBM hardware.
• 35:00Preliminary results: error profiles for temperature and velocity (1e‑6 to 1e‑5), and ongoing work to isolate quantum vs classical contributions.

• GitHub - AI_simulations — Repository containing code for AI simulations used in the talk.
• GitHub - quasi.ai — Repository for the Quasi AI framework.
• GitHub - qcpinn-convection-diffusion-qiskit — Code for quantum physics-informed neural networks for convection-diffusion using Qiskit.
• quasi.digital — Official website of Quasi AI, the startup presenting the research.
• Aditya Seshaditya LinkedIn — LinkedIn profile of the presenter.
• Qiskit Fall Fest 2026 — Event page for the Qiskit Fall Fest where the talk was hosted.
• Hackathon details — Information about the hackathon and IQM hardware credits.
• Seminar Series Playlist — Playlist of the seminar series in which this talk appears.