![[2602.15283] Complex-Valued Unitary Representations as Classification Heads for Improved Uncertainty Quantification in Deep Neural Networks](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.15283] Complex-Valued Unitary Representations as Classification Heads for Improved Uncertainty Quantification in Deep Neural Networks
Summary
This paper introduces a novel classification head architecture using complex-valued unitary representations to enhance uncertainty quantification in deep neural networks, demonstrating significant improvements in calibration metrics.
Why It Matters
As deep neural networks achieve high accuracy, their confidence scores often misrepresent true correctness probabilities. This research addresses the critical issue of model calibration, which is essential for applications in safety-critical domains like healthcare and autonomous systems. By leveraging complex-valued representations, the proposed method offers a promising approach to improve model reliability and interpretability.
Key Takeaways
- The proposed architecture uses complex-valued representations to improve calibration in neural networks.
- A significant reduction in Expected Calibration Error (ECE) was achieved compared to standard methods.
- The study highlights the potential of quantum-inspired techniques in enhancing model performance.
- Negative results in certain applications help define the method's limitations.
- Publicly available code facilitates further research and application of the findings.
Computer Science > Machine Learning arXiv:2602.15283 (cs) [Submitted on 17 Feb 2026] Title:Complex-Valued Unitary Representations as Classification Heads for Improved Uncertainty Quantification in Deep Neural Networks Authors:Akbar Anbar Jafari, Cagri Ozcinar, Gholamreza Anbarjafari View a PDF of the paper titled Complex-Valued Unitary Representations as Classification Heads for Improved Uncertainty Quantification in Deep Neural Networks, by Akbar Anbar Jafari and 2 other authors View PDF HTML (experimental) Abstract:Modern deep neural networks achieve high predictive accuracy but remain poorly calibrated: their confidence scores do not reliably reflect the true probability of correctness. We propose a quantum-inspired classification head architecture that projects backbone features into a complex-valued Hilbert space and evolves them under a learned unitary transformation parameterised via the Cayley map. Through a controlled hybrid experimental design - training a single shared backbone and comparing lightweight interchangeable heads - we isolate the effect of complex-valued unitary representations on calibration. Our ablation study on CIFAR-10 reveals that the unitary magnitude head (complex features evolved under a Cayley unitary, read out via magnitude and softmax) achieves an Expected Calibration Error (ECE) of 0.0146, representing a 2.4x improvement over a standard softmax head (0.0355) and a 3.5x improvement over temperature scaling (0.0510). Surprisingly, replacin...
