Physical mechanisms governing generalization and hallucination in deep learning for imaging through scattering media
), while hallucinations arise when this capacity is exceeded, resulting in unconstrained, non-physical predictions. We also identify residual ballistic light, if not negligible, as a stabilizing anchor, enabling robust predictions under scattering variability. Integrating experimental validation with wave-optics simulations, we establish a universal framework that links these phenomena, showing that strategic training on diverse physical mappings enhances generalization while suppressing hallucinations. This work bridges physics-driven interpretability with AI design, offering actionable strategies to develop reliable models for applications ranging from medical imaging through biological tissues to autonomous navigation in scattering environments.
Authors
- Haofan Huang (ORCID: https://orcid.org/0000-0002-3505-9972)
- Xuyu Zhang (ORCID: https://orcid.org/0000-0002-0637-3049)
- Puxiang Lai (ORCID: https://orcid.org/0000-0003-4811-2012)
- D Zhang
- Tianting Zhong (ORCID: https://orcid.org/0000-0002-3925-5903)
- Honglin Liu (ORCID: https://orcid.org/0000-0003-2473-9056)
- Shensheng Han
- Songlin Zhuang
Institutions
- Hong Kong Polytechnic University (HK)
- University of Shanghai for Science and Technology (CN)
- Shenzhen Polytechnic University (CN)
- Chinese Academy of Sciences (CN)
- Shanghai Institute of Optics and Fine Mechanics (CN)
- University of Chinese Academy of Sciences (CN)
Publication Details
- Journal
- Nature Communications
- Published
- 2026-04-23
- DOI
- https://doi.org/10.1038/s41467-026-72304-z
- Primary Topic
- Random lasers and scattering media
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China
- Innovation and Technology Commission
- Hong Kong Polytechnic University
- Science, Technology and Innovation Commission of Shenzhen Municipality