Skip to content

Finals hardening - #2

Merged
ravindraogg merged 2 commits into
mainfrom
finals-hardening
Jul 26, 2026
Merged

ravindraogg merged 2 commits into
mainfrom
finals-hardening

Conversation

@ravindraogg

Copy link
Copy Markdown
Member

No description provided.

Trivikram-Narayanan and others added 2 commits July 26, 2026 00:28
…es the failed Qiskit VQC

The Quant-A-thon primer teaches both Qiskit and PennyLane and points to PennyLane
for QML (differentiable circuits). The project was Qiskit-only, and its Qiskit VQC
(gradient-free COBYLA/SPSA) underperformed (ROC-AUC 0.50-0.63, near random).

radarsense/vqc_pennylane.py trains the same ansatz (AngleEmbedding +
StronglyEntanglingLayers + expval) with true GRADIENT DESCENT (Adam), same 4
features / data / seeds. Result: ROC-AUC 0.979 +/- 0.004, PR-AUC 0.924 --
competitive with classical (0.979 / 0.939). So the earlier VQC failure was an
optimizer problem, not a fundamental one -- exactly the primer's slide-12 point.

Honest: this does NOT overturn the no-quantum-advantage conclusion (classical
still edges the VQC); it turns a broken VQC into a working one. Device-agnostic
verified (slide 14): identical output on default.qubit and lightning.qubit;
qiskit.aer/IBM is a one-line device swap. Both SDKs now used.
@ravindraogg
ravindraogg merged commit bc603d6 into main Jul 26, 2026
0 of 2 checks passed
Trivikram-Narayanan pushed a commit that referenced this pull request Aug 12, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants