AI researchinfo
Exact Born evidence maps for quantum and optical image classifiers
Original title: Seeing Why: Exact Evidence Maps for Quantum and Optical Image Classifiers
AI summary · x-ai/grok-4.7Summaries by Vani AI
Deleting highest evidence pixels drops accuracy from 73% to 9% while gradient saliency removes less.
The paper presents exact evidence maps for quantum and optical classifiers using amplitude-encoded images. Born evidence maps decompose class probabilities exactly into amplitude contributions. Experiments on IBM hardware show 37.7% accuracy vs 10% chance on Caltech-10.
Key facts
- Accuracy drops from 73% to 9% when deleting 5% highest evidence pixels
- Fourier-plane mask followed by brickwall circuit on qubits
- 37.7% accuracy on IBM ibm_kingston for Caltech-10
- Maps reveal data shortcuts
From the original article
1 min read at arxiv.orgWhen an image classifier decides, we can usually only estimate why.
By Teerawat Tongloy · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.