Mooseok Jang

As AI capabilities advance, AI reviewers (agents that read a manuscript and produce a referee report) are beginning to be deployed in scientific peer review. Yet, their credibility remains in question: many scientists think that they cannot judge scientific validity, while others are more optimistic about their readiness. In this work, we conduct a study in which 45 scientists in Physical, Biological, and Health Sciences spend 469 hours rating 2,960 individual criticisms (each targeting one aspect of a paper) from human-written and AI-generated reviews of 82 Nature-family papers, scoring each on correctness, significance, and evidence sufficiency. On a composite rating of reviews across all three dimensions, an AI reviewer powered by GPT-5.2 scores above each paper's top-rated human reviewer (60.0% vs 48.2% of criticisms, p=0.009). Also, AI reviewers' correct criticisms are more often rated significant, and surface a distinct 26% of legitimate issues that no human raises. However, AI reviewers are more often incorrect, overlap with each other far more than humans do with each other, and exhibit 16 weaknesses rare in human reviews. Overall, our results position current AI reviewers as complements to, not substitutes for, human reviewers.

Interferometric displacement sensing is widely used for non-contact precision measurement because coherent phase readout resolves extremely small dynamic displacements. However, in conventional OPD-modulation interferometers, the displacement-to-phase responsivity is primarily defined by the optical wavelength and interrogation geometry, providing limited freedom to independently scale the phase excursion for a given displacement. Here, we demonstrate a scalable phase-sensitive interferometric displacement sensor (SPIDS) that circumvents this constraint through external-cavity-laser (ECL)-mediated phase transduction. Target displacement perturbs the ECL cavity condition and induces a lasing-wavenumber shift, which a delay-interferometric readout converts into an enlarged interferometric phase response, so that the responsivity scales with the delay-to-cavity-length ratio rather than with the optical wavelength alone. The measured phase response increased linearly with the added fiber delay length, reaching an approximately 965-fold enhancement at a 300 m delay relative to a wavelength-fixed homodyne baseline under identical piezoelectric actuation. The enlarged responsivity further enabled reconstruction of an estimated sub-nanometer dynamic displacement under weak actuation and non-contact detection of photoacoustically induced transient surface displacement. These results establish ECL-mediated phase transduction as an externally scalable displacement-to-phase conversion mechanism, providing increased phase-readout margin for weak dynamic and transient displacement sensing.

Here, we present a pupil-modulation method that selectively attenuates the scattered field from an object, enabling quantitatively accurate phase retrieval based on spatial-domain Kramers-Kronig (KK) relations. Numerical simulations show that conventional KK reconstruction suffers from increasing phase errors when the weak-scattering assumptionwhich requires the detected field to be dominated by ballistic light-is violated due to increased phase contrast, amplitude variation, or spatial phase complexity. The proposed method is experimentally validated on two representative reflective samples: one with uniform reflectivity and another exhibiting amplitude variations. The retrieved phase maps demonstrate close agreement with both theoretical predictions and off-axis interferometric measurements. This work establishes a practical pathway for extending KK-based quantitative phase imaging (QPI) to strongly scattering and reflective surfaces, offering a non-interferometric, deterministic, and non-iterative tool for metrology applications with minimal computational overhead.

Spectral sensing provides critical insight into the properties of materials or structures, yet conventional spectrometers face fundamental limitations in simultaneously achieving high spectral resolution, compact form factors, robustness to environmental noise, and simplified calibration processes. Recently, we reported a double-layer disordered metasurface platform as a breakthrough to overcome such conventional trade-offs [Lee et al. (2025)]. This disordered dispersive element enables the engineering of multiple scattering for highly sensitive spectral-to-spatial encoding. With this predictable spectral-to-spatial mixing feature, we demonstrate a reconstructive spectrometer with a 1-nm-scale spectral resolution over the broad visible spectrum from 440 nm to 660 nm in a sub centimeter form factor. Here, we highlight this breakthrough along with relevant studies in the broader area of reconstructive spectroscopy, particularly those published in 2025. We envision that the stacked disordered metasurface platform can serve as a universal linear optical encoder.

Inline holographic imaging presents an ill-posed inverse problem of reconstructing the complex amplitude of the scattered wavefield, which is governed by the object’s complex refractive index, from recorded diffraction patterns. Although recent deep learning approaches have shown promise over classical phase retrieval algorithms, they often require high-quality ground truth datasets of complex amplitude maps to achieve a statistical inverse mapping operation between the two domains. Here, a physics-aware style transfer framework is proposed that treats the object-to-sensor distance as an implicit style within diffraction patterns. Using the style domain as the intermediate domain to construct cyclic image translation, inverse mapping can be learned in an adaptive manner only with datasets composed of intensity measurements. Biomedical applicability is demonstrated by reconstructing the morphology of dynamically flowing red blood cells, highlighting its potential for real-time, label-free imaging. As a framework that leverages physical cues inherently embedded in measurements, the presented method offers a practical learning strategy for imaging applications where ground truth is difficult or impossible to obtain.

Polarized on-chip lensless microscopy provides a cost-effective diagnostic solution for large field-of-view (FOV) imaging. However, conventional designs often rely on a 3-dimensional and rotational motorized stage for the sensor, a configuration that necessitates an additional image registration process, increases system bulkiness, and extends image acquisition time. In this work, we present a polarized lensless microscope based on illumination diversity. The proposed system integrates a two-LED multiplexed pixel super-resolution and multi-wavelength phase retrieval framework based on an RGB-LED matrix, as well as illumination polarization-state modulation, enabling high-resolution polarized microscopic imaging without mechanical movement in the detection optics, including the specimen. Our system achieves sub-micron resolution across a wide FOV while featuring a compact, cost-effective design with rapid image acquisition and reconstruction. We expect that our proposed polarized lensless microscope design will enhance the applicability of lensless imaging systems in the biomedical field.

Fourier ptychography microscope (FPM) is a computational imaging technique that can achieve a large field‐of‐view and high‐resolution retrieval of complex‐amplitude object function from a set of low‐resolution intensity images. This study investigates the effects of system‐model discrepancies in FP reconstruction under diverse scenarios characterized by experimental errors, including axial mispositioning of specimens, sample thickness, and factors affecting k‐vector sampling positions—magnification, light‐emitting diode (LED) height, and positional errors of LED elements. With the systematic investigation using experimental data and Gauss‐Newton algorithm, we provided the level of tolerance for different types of experimental errors and its general relation to configurational parameters for FPM, such as the numerical aperture of objective lens, synthetic numerical aperture, and the spectrum overlapping ratio of low‐resolution images. Also, we found that different types of errors can generate characteristic effects on the appearance of the reconstructed images, and therefore, reconstruction results could be highly inconsistent under various conditions of system‐model mismatch, even if the reconstructed images are seemingly plausible. The investigation highlights the need for an improved algorithmic approach to physically accurate retrieval of complex‐amplitude maps using a FPM.

Imaging through scattering is challenging, as even a thin layer can randomly perturb light propagation and obscure hidden objects. Accurate closed-form modeling of forward scattering remains difficult, particularly for dynamically varying or thick layers. Here, we introduce a plug-and-play inverse solver based on video diffusion models with a physically grounded forward model tailored to dynamic scattering layers. Our method extends Diffusion Posterior Sampling (DPS) to the spatio-temporal domain, thereby capturing statistical correlations between video frames and scattered signals more effectively. Leveraging these temporal correlations, our approach recovers high-resolution spatial details that spatial-only methods typically fail to reconstruct. We also propose an inference-time optimization with a lightweight mapping network, enabling joint estimation of low-dimensional forward-model parameters without additional training. This joint optimization significantly enhances adaptability to unknown, time-varying degradations, making our method suitable for blind inverse scattering problems. We validate across diverse conditions, including different scene types, layer thicknesses, and scene-layer distances. And real-world experiments using multiple datasets confirm the robustness and effectiveness of our approach, even under real noise and forward-model approximation mismatches. Finally, we validate our method as a general video-restoration framework across dehazing, deblurring, inpainting, and blind restoration under complex optical aberrations.

Inline holographic imaging presents an ill-posed inverse problem of reconstructing objects' complex amplitude from recorded diffraction patterns. Although recent deep learning approaches have shown promise over classical phase retrieval algorithms, they often require high-quality ground truth datasets of complex amplitude maps to achieve a statistical inverse mapping operation between the two domains. Here, we present a physics-aware style transfer approach that interprets the object-to-sensor distance as an implicit style within diffraction patterns. Using the style domain as the intermediate domain to construct cyclic image translation, we show that the inverse mapping operation can be learned in an adaptive manner only with datasets composed of intensity measurements. We further demonstrate its biomedical applicability by reconstructing the morphology of dynamically flowing red blood cells, highlighting its potential for real-time, label-free imaging. As a framework that leverages physical cues inherently embedded in measurements, the presented method offers a practical learning strategy for imaging applications where ground truth is difficult or impossible to obtain.