Bayram Eker
Menu
EN
CV Start a conversation

Home / Writing / Systems

Why I keep the planner classical

A quantum component can help with a calculation without taking over the decision. Keeping that distinction clear makes QANTIS easier to test and use.

Uncrewed assurance console showing decision evidence and authority state
A recommendation reaches the person responsible for the next step. NODERIQ console concept.

When a new component looks promising, it is natural to imagine what it could do at the centre of a system. I also want to imagine the day it is unavailable. If the quantum processor cannot answer, what happens to the decision waiting for it? That question helps explain a choice in both QANTIS papers: probabilities are updated by the quantum component, while a classical planner chooses the action.

This is a design choice I intend to keep studying, rather than a promise to move everything to quantum hardware later. A defined calculation is easier to compare, replace and understand when its limits are visible.

The system needs a way to keep going

Remote quantum backends have queues and calibration periods, and they can be offline. A classical path lets the planner continue when the research component cannot help. It also gives us a direct reference for asking whether the extra component contributed anything.

Classical does not mean infallible. The planner still needs tests and clear conditions for use. The useful point is that we can examine its behaviour separately from the experimental calculation, rather than trying to explain both at once.

A small component gives us a better experiment

The belief-update input and output are defined distributions. That lets us compare against exact Bayes and follow repeated use, as the sequential study does over 8, 12, 20 and 32 steps. When something changes, we have a clearer chance of locating the cause.

The component can also be removed. If the quantum path never meets the G3 whole-workflow test, NODERIQ’s classical core should still be useful. An experiment should be allowed to disappoint us without taking the programme’s basic purpose with it.

There is a real trade-off

A tightly defined role may leave some possibilities unexplored. I take that objection seriously. For this research, I prefer a contribution we can measure to a larger arrangement whose effects are difficult to separate. It gives us a result another team can question and test.

The same idea appears in the handover to a person. A calculation can inform a recommendation without granting permission to act. In NODERIQ, I want the recommendation and its basis to reach the person responsible for the next step. Keeping the planner classical is one concrete place to practise that separation.

A calculation inside a decision
  • Update with the experimental component
  • Compare with the classical reference
  • Keep the planner and fallback available

Sources

  1. QANTIS: A Hardware-Validated Quantum Platform for POMDP Planning and Multi-Target Data Association (preprint) arxiv.org
  2. QANTIS: Hardware-Calibrated Sequential POMDP Belief Updates on IBM Heron (preprint) arxiv.org

Written by Bayram Yüksel Eker. Corrections and disagreement are welcome at bayram@neuraparse.com.