George Barbastathis

We show that a bounded temporal increment prior on the sample dynamics is sufficient to reconstruct a time-varying phase object from a near-field diffraction movie, under the thin-film approximation. The time evolution of the field is parameterized by a multiplicative inter-frame update factor, and a bound on its complex-plane support enforces a bounded phase increment and a passive amplitude constraint. Reconstruction of the dynamic field is thereby converted into a feasibility problem with two projection operators: a measurement-domain modulus projection and an object-domain circular-sector projection. We validate the approach experimentally using a spatial light modulator as a calibrated dynamic sample in two cases: a reaction--diffusion phase pattern with spatially expanding extent, and a growing phase pattern whose accumulated phase reaches $10π$. In both cases the reconstructed phase trajectory agrees well with the ground truth. We then apply the same framework, without modification, to in-situ monitoring of a photo-polymer 3D printing process, recovering the spatiotemporal phase induced by polymerization under a spatially patterned blue light. The reconstructed phase trajectory provides an observable for photo-chemical system identification and process control.

We show that a bounded temporal increment prior on the sample dynamics is sufficient to reconstruct a time-varying phase object from a near-field diffraction movie, under the thin-film approximation. The time evolution of the field is parameterized by a multiplicative inter-frame update factor, and a bound on its complex-plane support enforces a bounded phase increment and a passive amplitude constraint. Reconstruction of the dynamic field is thereby converted into a feasibility problem with two projection operators: a measurement-domain modulus projection and an object-domain circular-sector projection. We validate the approach experimentally using a spatial light modulator as a calibrated dynamic sample in two cases: a reaction--diffusion phase pattern with spatially expanding extent, and a growing phase pattern whose accumulated phase reaches $10π$. In both cases the reconstructed phase trajectory agrees well with the ground truth. We then apply the same framework, without modification, to in-situ monitoring of a photo-polymer 3D printing process, recovering the spatiotemporal phase induced by polymerization under a spatially patterned blue light. The reconstructed phase trajectory provides an observable for photo-chemical system identification and process control.

Rapid and robust molecular fingerprinting is critical in biomanufacturing, diagnostics, and environmental monitoring. Nanopore sensing provides single-molecule readouts as transient ionic current pulses; however, conventional analyses depend on handcrafted features that miss informative structural information. We present an interpretable machine learning framework that operates directly on raw pulses, pairing a physics-guided time-frequency transform with a compact neural classifier and feature-attribution maps. We also include conventional feature-based SVMs and a 1D classifier trained on raw pulses as baselines. On two self-assembled DNA nanostructures of similar size but distinct geometry, for which standard pulse features overlap, the method achieves high accuracy and yields physically consistent attributions that highlight discriminative signal motifs. A matched control without the time-frequency transform clarifies when learned filters suffice versus when physics-guided preprocessing improves reliability, leading to a practical "custom-filter" design principle. The workflow is modular, lightweight, and applicable to pulse-based sensing platforms, including virus and exosome analysis, electrochemical monitoring, and industrial fault detection. By combining accuracy with transparency, it lays the groundwork for deployable sensing platforms in regulated, mission-critical settings.

Purpose: The purposes of this study were to assess whether optic nerve head (ONH) biomechanics, quantified by tissue strain, improves classification of progressive visual field (VF) loss patterns in glaucoma beyond morphology, and to use saliency maps to identify ONH regions associated with the predictions. Methods: We recruited 249 patients with glaucoma (mean age 69 ± 5 years, 54% female patients). One eye per subject was imaged under (1) primary gaze and (2) primary gaze with IOP elevated to approximately 35 millimeters of mercury (mm Hg) via ophthalmo-dynamometry. Twelve subjects were excluded due to poor scan quality/limited lamina cribrosa (LC) visibility. Experts classified subjects into four categories based on the presence of specific visual field defects (VFDs): (1) superior nasal step (N = 26), (2) superior partial arcuate (N = 62), (3) full superior hemifield defect (N = 25), and (4) other/non-specific defects (N = 124). Automatic segmentation and digital volume correlation computed neural tissue and LC strains. Biomechanical and structural features were input to a PointNet model. Three classification tasks were performed to detect: (1) superior nasal step, (2) superior partial arcuate, and (3) full superior hemifield defect. Data were split 80/20 (train/test). Area under the curve (AUC) assessed performance. Saliency maps (an explainable artificial intelligence [XAI] technique) highlighted ONH regions most critical to classification. Results: Models achieved AUCs of 0.77 to 0.88 across VFD classifications. The structure-only model reached an AUC of 0.83 ± 0.02 for superior arcuate defects, which significantly improved to 0.87 ± 0.02 (P < 0.05) with the addition of strain information, demonstrating that ONH biomechanics enhance prediction beyond morphology. Strain-sensitive regions were localized to the inferior and inferotemporal rim, expanding with increasing severity of VF loss. Conclusions: ONH strain enhances classification of glaucomatous VF loss patterns. The neuroretinal rim, rather than the LC, was most critical, suggesting rim strain may play a dominant role in axonal injury and functional loss.

Understanding nanoscale kinetic processes during electrochemical deposition is critical for advancing a range of technologies such as heterogeneous catalysis, microelectronics manufacturing, energy storage systems and precision fabrication of functional materials. Copper pulse-reversal electrodeposition (Cu PR-ED) serves as an ideal model system to probe these dynamics, where control over deposition morphology through parameters like current density and pulse timing remains empirically established but mechanistically unresolved. This study employs operando synchrotron X-ray nanoimaging combined with multiscale characterization to resolve interfacial processes at previously inaccessible sub-10nm and millisecond spatiotemporal resolutions. A key innovation lies in overcoming the inherent trade-off between imaging fidelity and beam-sample interactions for radiation sensitive materials: our methodology, which uses computational and machine learning-based denoising and background correction, enables the analysis of images acquired under reduced X-ray doses. This approach allows for observation of intrinsic deposition phenomena while minimizing radiation artifacts under operando conditions. Comparative growth analysis under different attenuated radiation levels reveals distinct kinetic regimes governed by interfacial transport conditions, with implications for nucleation and growth mechanisms across electrochemical systems. Beyond advancing Cu PR-ED control, this work establishes a general framework for studying dynamic processes in beam-sensitive materials, directly addressing fundamental limitations identified in radiation-based nanoscale imaging.

The surge in artificial intelligence applications calls for scalable, high-speed, and low-energy computation methods. Computing with photons is promising due to the intrinsic parallelism, high bandwidth, and low latency of photons. However, current photonic computing architectures are limited by the speed and energy consumption associated with electronic-to-optical data transfer, i.e., electro-optic conversion. Here, we demonstrate a thin-film lithium niobate (TFLN) computing circuit that addresses this challenge, leveraging both highly efficient electro-optic modulation and the spatial scalability of TFLN photonics. Our circuit is capable of computing at 43.8 GOPS/channel while consuming 0.0576 pJ/OP, and we demonstrate various inference tasks with high accuracy, including the classification of binary data and complex images. Heightening the integration level, we show another TFLN computing circuit that is combined with a hybrid-integrated distributed-feedback laser and heterogeneous-integrated modified uni-traveling carrier photodiode. Our results show that the TFLN photonic platform holds promise to complement silicon photonics and diffractive optics for photonic computing, with extensions to ultrafast signal processing and ranging. Efficient electro-optic conversion is central to photonic computing, and thin-film lithium niobate (TFLN) offers this capability. Here, the authors demonstrate computing circuits on the TFLN platform, enabling the next generation of photonic computing systems featuring both high-speed and low-power.

We present a coupled nonlinear optimization framework that combines a physics-based dynamical model K with an optical propagation model H to perform dynamic low-dose characterization directly from time-resolved diffraction data. By embedding the equations of motion within the optical forward operator, we allow for mutual regularization between imaging and dynamics. The method converts the inverse imaging problem into a parameter-estimation task, thereby avoiding frame-by-frame phase retrieval and suppressing ill-conditionedness arising from noisy data. The approach is validated on synchrotron X-ray movies of a thermally actuated micro-electro-mechanical (MEMS) oscillator. At a photon dose of ≈ 4 photons per pixel per frame, it simultaneously reconstructs the shuttle edge profile, partially coherent probe modes, and their temporal occupancies. With an incident flux of ≈ 0.015 photons per pixel per frame, the framework still recovers the MEMS shuttle displacement trajectory. Compared with conventional two-step pipelines, the joint treatment yields improved noise robustness by exploiting temporal correlations and enforcing physically admissible motion.

Two-color projection micro-stereolithography (PμSL) is an advanced additive manufacturing technique that enables rapid and continuous 3D printing. This method uses two distinct wavelengths of light to independently control photoinitiation and photoinhibition of polymerization in a photocurable resin, achieved by selecting a photoinitiator and photoinhibitor with complementary absorption spectra, enabling high-precision, continuous printing compatible with viscous resins. PμSL has significant potential in applications such as the fabrication of optical diffractive neural networks (D2NNs), which require sophisticated three-dimensional photonic structures capable of optical computation. However, achieving high-precision 3D printing with PμSL remains challenging due to limitations in current inspection and modeling techniques. The photochemical processes involved in photopolymerization are highly complex, with unknown local interactions among different species and non-local diffusion effects arising from spatial concentration differences in the reaction region. Accurate modeling of this lithography process is necessary for inverse design to fabricate precise sub-diffraction-limit polymer features. In this study, we develop a dynamic model for the polymerization process during 3D printing. We first establish a local generalized Lotka-Volterra system with parameters estimated from Fourier Transform Infrared Spectroscopy (FT-IR) measurements. We then extend this system to a reaction-diffusion model with unknown diffusion coefficients. To address residuals not captured by the symbolic models, we incorporate a neural network as a universal function approximator. Simulation results align closely with imaging measurements, and further analysis demonstrates improved generalization when additional physical priors are applied to the diffusion coefficients.

Lyophilization, aka freeze drying, is a process commonly used to increase the stability of various drug products in biotherapeutics manufacturing, e.g., mRNA vaccines, allowing for higher storage temperature. While the current trends in the industry are moving towards continuous manufacturing, the majority of industrial lyophilization processes are still being operated in a batch mode. This article presents a framework that accounts for the probabilistic uncertainty during the primary and secondary drying steps in continuous lyophilization. The probabilistic uncertainty is incorporated into the mechanistic model via polynomial chaos theory (PCT). The resulting PCT-based model is able to accurately and efficiently quantify the effects of uncertainty on several critical process variables, including the temperature, sublimation front, and concentration of bound water. The integration of the PCT-based model into stochastic optimization and control is demonstrated. The proposed framework and case studies can be used to guide the design and control of continuous lyophilization while accounting for probabilistic uncertainty.

This article describes a new, efficient way of finding control and state trajectories in optimal control problems by reformulation as a system of differential-algebraic equations (DAEs). The optimal control and state vectors can be obtained via simulation of the resulting DAE system with the selected DAE solver, eliminating the need for an optimization solver. Our simulation-based approach is demonstrated and benchmarked against various optimization-based algorithms via four case studies associated with the optimization and control of a Stefan problem for cell therapy. The simulation-based approach is faster than every optimization-based method by more than an order of magnitude while giving similar/better accuracy in all cases. The solution obtained from the simulation-based approach is guaranteed to be optimal provided that at least one constraint or algebraic equation resulting from the reformulation remains active at all times. The proposed technique offers an efficient and reliable framework for optimal control, serving as a promising alternative to the traditional techniques in applications where speed is crucial, e.g., real-time online model predictive control.

Phenomena involving multiple scattering, despite having attracted considerable attention in physics for decades, continue to generate unexpected and counterintuitive behaviours prompting further studies. For optical scattering, the memory effect well predicts fourth order statistics, i.e. the intensity correlation, as long as the scattering strength and depth are within certain bounds. The memory effect has found a wide range of applications, where its limitations also become apparent: for example, in imaging through turbid media, decorrelation due to multiscattering in thick samples has been shown to restrict the field of view. However, to our knowledge, no comprehensive mechanism exists to date that can account for decorrelation precisely. In this paper, we quantify how the scatterer's own statistics determine such limitations. We show that the ensemble statistics of the backscattered field may be decomposed into two terms: one expresses surface scattering, where statistical distributions of multiscale structure features may be inferred from our previous works; while the second term originates from the underlying scattering volume and is diffusive. The new framework agrees well with experiments, including the prediction of a new quasipower law for fluctuations induced by the single realization.

Lyophilization, aka freeze drying, is a process commonly used to increase the stability of various drug products in biotherapeutics manufacturing, e.g., mRNA vaccines, allowing for higher storage temperature. While the current trends in the industry are moving towards continuous manufacturing, the majority of industrial lyophilization processes are still being operated in a batch mode. This article presents a framework that accounts for the probabilistic uncertainty during the primary and secondary drying steps in continuous lyophilization. The probabilistic uncertainty is incorporated into the mechanistic model via polynomial chaos theory (PCT). The resulting PCT-based model is able to accurately and efficiently quantify the effects of uncertainty on several critical process variables, including the temperature, sublimation front, and concentration of bound water. The integration of the PCT-based model into stochastic optimization and control is demonstrated. The proposed framework and case studies can be used to guide the design and control of continuous lyophilization while accounting for probabilistic uncertainty.

We develop a method that reconstructs nano-oscillator displacement from low-dose diffraction movies by jointly optimizing over dynamics and optical model; the formulation readily extends to simultaneously retrieve unknown sample shape and probe.

Revised Supplement

Supplemental material

Supplemental material

Revised Supplement

This article describes a new, efficient way of finding control and state trajectories in optimal control problems by reformulation as a system of differential-algebraic equations (DAEs). The optimal control and state vectors can be obtained via simulation of the resulting DAE system with the selected DAE solver, eliminating the need for an optimization solver. Our simulation-based approach is demonstrated and benchmarked against various optimization-based algorithms via four case studies associated with the optimization and control of a Stefan problem for cell therapy. The simulation-based approach is faster than every optimization-based method by more than an order of magnitude while giving similar/better accuracy in all cases. The solution obtained from the simulation-based approach is guaranteed to be optimal provided that at least one constraint or algebraic equation resulting from the reformulation remains active at all times. The proposed technique offers an efficient and reliable framework for optimal control, serving as a promising alternative to the traditional techniques in applications where speed is crucial, e.g., real-time online model predictive control.

PURPOSE (1) To assess whether neural tissue structure and biomechanics could predict functional loss in glaucoma; (2) To evaluate the importance of biomechanics in making such predictions. DESIGN Clinic-based cross-sectional study. METHODS We recruited 238 glaucoma subjects (Chinese ethnicity, more than 50 years old). For one eye of each subject, we imaged the optic nerve head (ONH) using spectral-domain OCT under the following conditions: (1) primary gaze and (2) primary gaze with acute IOP elevation (to approximately 35 mmHg) achieved through ophthalmo-dynamometry. We utilized automatic segmentation of optic nerve head (ONH) tissues and digital volume correlation (DVC) analysis to compute intraocular pressure (IOP)-induced neural tissue strains. A robust geometric deep learning approach, known as Point-Net, was employed to predict the full Humphrey 24-2 pattern standard deviation (PSD) maps from ONH structural and biomechanical information. For each point in each PSD map, we predicted whether it exhibited no defect or a PSD value of less than 5%. Predictive performance was evaluated using 5-fold cross-validation and the F1-score. We compared the model's performance with and without the inclusion of IOP-induced strains to assess the impact of biomechanics on prediction accuracy. RESULTS Integrating biomechanical (IOP-induced neural tissue strains) and structural (tissue morphology and neural tissues thickness) information yielded a significantly better predictive model (F1-score: 0.76 ± 0.02) across validation subjects, as opposed to relying only on structural information, which resulted in a significantly lower F1-score of 0.71 ± 0.02 (p < 0.05). Our subjects had a mean age of 69±5 years. Among them, 88 were female. The cohort included a wide range of glaucoma severity, with Mean Deviation (MD) values ranging from -1.8 (mild) to -25.2 (severe), and an average MD value of -7.25±5.05. CONCLUSION Our study has shown that the integration of biomechanical data can significantly improve the accuracy of visual field loss predictions and highlights the importance of the biomechanics-function relationship in glaucoma.

Here we show a photonic computing accelerator utilizing a system-level thin-film lithium niobate circuit which overcomes this limitation. Leveraging the strong electro-optic (Pockels) effect and the scalability of this platform, we demonstrate photonic computation at speeds up to 1.36 TOPS while consuming 0.057 pJ/OP. Our system features more than 100 thin-film lithium niobate high-performance components working synergistically, surpassing state-of-the-art systems on this platform. We further demonstrate binary-classification, handwritten-digit classification, and image classification with remarkable accuracy, showcasing our system's capability of executing real algorithms. Finally, we investigate the opportunities offered by combining our system with a hybrid-integrated distributed feedback laser source and a heterogeneous-integrated modified uni-traveling carrier photodiode. Our results illustrate the promise of thin-film lithium niobate as a computational platform, addressing current bottlenecks in both electronic and photonic computation. Its unique properties of high-performance electro-optic weight encoding and conversion, wafer-scale scalability, and compatibility with integrated lasers and detectors, position thin-film lithium niobate photonics as a valuable complement to silicon photonics, with extensions to applications in ultrafast and power-efficient signal processing and ranging.

To study biological signalling, great effort goes into designing sensors whose fluorescence follows the concentration of chemical messengers as closely as possible. However, the binding kinetics of the sensors are often overlooked when interpreting cell signals from the resulting fluorescence measurements. We propose a method to reconstruct the spatiotemporal concentration of the underlying chemical messengers in consideration of the binding process. Our method fits fluorescence data under the constraint of the corresponding chemical reactions and with the help of a deep-neural-network prior. We test it on several GCaMP calcium sensors. The recovered concentrations concur in a common temporal waveform regardless of the sensor kinetics, whereas assuming equilibrium introduces artifacts. We also show that our method can reveal distinct spatiotemporal events in the calcium distribution of single neurons. Our work augments current chemical sensors and highlights the importance of incorporating physical constraints in computational imaging. A key aspect of biosensor design is ensuring that fluorescent signals follow the concentration of the analytes as closely as possible, but binding kinetics are often overlooked. Here authors propose a method for reconstructing the spatiotemporal concentration of the underlying chemical messengers by considering the binding process.

The quantitative phase image formation process is posed as a problem of parameter estimation from intensity measurements. This approach is inclusive of traditional pixel-oriented imaging, where the sought parameters are the pixel values. The resulting optimization process to find the parameters is then seen to depend on the sensitivity field: this is the gradient of the scattered field with respect to the parameters, and it turns out to obey a scattering relationship that is analogous to that of the original scattered field. Examples are given from several regimes of scattering strength.

Lyophilization (aka freeze drying) has been shown to provide long-term stability for many crucial biotherapeutics, e.g., mRNA vaccines for COVID-19, allowing for higher storage temperature. The final stage of lyophilization, namely secondary drying, entails bound water removal via desorption, in which accurate prediction of bound water concentration is vital to ensuring the quality of the lyophilized product. This article proposes a novel technique for real-time estimation of the residual moisture during secondary drying in lyophilization. A state observer is employed, which combines temperature measurement and mechanistic understanding of heat transfer and desorption kinetics, without requiring any online concentration measurement. Results from both simulations and experimental data show that the observer can accurately estimate the concentration of bound water in real time for all possible concentration levels, operating conditions, and measurement noise. This framework can also be applied for monitoring and control of the residual moisture in other desorption-related processes.

Abstract Optimal control over fast chemical processes hinges on the achievement of rapid and effective mixing. Impinging jet mixers are a unique class of passive mixing devices renowned for their exceptional ability to achieve rapid mixing at micro‐length scales, whilst offering the possibility of a high throughput. Comprising of two co‐linear jets flowing in opposite directions and colliding with each other within a small (usually confined) volume, these devices effectively intensify various mixing‐controlled processes in a reproducible manner. Impinging jet mixers find extensive use in both the chemical and pharmaceutical industry for a plethora of applications, such as reaction injection molding and precipitation processes. This review provides an overview of research related to impinging jet mixers, with an emphasis on the mixing characteristics and the influence of design and process parameters on the mixing performance. Lastly, specific applications for which these devices are exceptionally suited are discussed.

Non-invasive characterization of powders may take one of two approaches: imaging and counting individual particles; or relying on scattered light to estimate the particle size distribution (PSD) of the ensemble. The former approach runs into practical difficulties, as the system must conform to the working distance and other restrictions of the imaging optics. The latter approach requires an inverse map from the speckle autocorrelation to the particle sizes. The principle relies on the pupil function determining the basic sidelobe shape, whereas the particle size spread modulates the sidelobe intensity. We recently showed that it is feasible to invert the speckle autocorrelation and obtain the PSD using a neural network, trained efficiently through a physics-informed semi-generative approach. In this work, we eliminate one of the most time-consuming steps of our previous method by engineering the pupil function. By judiciously blocking portions of the pupil, we sacrifice some photons but in return we achieve much enhanced sidelobes and, hence, higher sensitivity to the change of the size distribution. The result is a 60 × reduction in total acquisition and processing time, or 0.25 seconds per frame in our implementation. Almost real-time operation in our system is not only more appealing toward rapid industrial adoption, it also paves the way for quantitative characterization of complex spatial or temporal dynamics in drying, blending, and other chemical and pharmaceutical manufacturing processes.

Lyophilization (aka freeze drying) has been shown to provide long-term stability for many crucial biotherapeutics, e.g., mRNA vaccines for COVID-19, allowing for higher storage temperature. The final stage of lyophilization, namely secondary drying, entails bound water removal via desorption, in which accurate prediction of bound water concentration is vital to ensuring the quality of the lyophilized product. This article proposes a novel technique for real-time estimation of the residual moisture during secondary drying in lyophilization. A state observer is employed, which combines temperature measurement and mechanistic understanding of heat transfer and desorption kinetics, without requiring any online concentration measurement. Results from both simulations and experimental data show that the observer can accurately estimate the concentration of bound water in real time for all possible concentration levels, operating conditions, and measurement noise. This framework can also be applied for monitoring and control of the residual moisture in other desorption-related processes.

This article describes a technique for solving optimal control problems by transformation into a system of differential-algebraic equations (DAEs). The optimal control vector can be obtained via simulation of the resulting DAE system with the selected DAE solver, eliminating the need for an optimization solver. This simulation-based (DAE-based) technique is demonstrated and benchmarked against various optimization-based approaches via two case studies associated with optimization and control of a Stefan problem. Results show that the simulation-based approach is faster than every optimization-based method by more than an order of magnitude while giving accurate solutions in all cases. The proposed framework offers a promising alternative to the traditional techniques in optimal control-related applications where speed is crucial, e.g., real-time online model predictive control.

The quantitative phase image formation process is posed as a problem of parameter estimation from intensity measurements. This approach is inclusive of traditional pixel-oriented imaging, where the sought parameters are the pixel values. The resulting optimization process to find the parameters is then seen to depend on the Sensitivity Field: this is the gradient of the scattered field with respect to the parameters, and it turns out to obey a scattering relationship that is analogous to that of the original scattered field. Examples are given from several regimes of scattering strength.

Objective. (1) To assess whether neural tissue structure and biomechanics could predict functional loss in glaucoma; (2) To evaluate the importance of biomechanics in making such predictions. Design, Setting and Participants. We recruited 238 glaucoma subjects. For one eye of each subject, we imaged the optic nerve head (ONH) using spectral-domain OCT under the following conditions: (1) primary gaze and (2) primary gaze with acute IOP elevation. Main Outcomes: We utilized automatic segmentation of optic nerve head (ONH) tissues and digital volume correlation (DVC) analysis to compute intraocular pressure (IOP)-induced neural tissue strains. A robust geometric deep learning approach, known as Point-Net, was employed to predict the full Humphrey 24-2 pattern standard deviation (PSD) maps from ONH structural and biomechanical information. For each point in each PSD map, we predicted whether it exhibited no defect or a PSD value of less than 5%. Predictive performance was evaluated using 5-fold cross-validation and the F1-score. We compared the model's performance with and without the inclusion of IOP-induced strains to assess the impact of biomechanics on prediction accuracy. Results: Integrating biomechanical (IOP-induced neural tissue strains) and structural (tissue morphology and neural tissues thickness) information yielded a significantly better predictive model (F1-score: 0.76+-0.02) across validation subjects, as opposed to relying only on structural information, which resulted in a significantly lower F1-score of 0.71+-0.02 (p &lt; 0.05). Conclusion: Our study has shown that the integration of biomechanical data can significantly improve the accuracy of visual field loss predictions. This highlights the importance of the biomechanics-function relationship in glaucoma, and suggests that biomechanics may serve as a crucial indicator for the development and progression of glaucoma.

Machine learning operators, such as neural networks, are universal function approximators—albeit, in practice, their generalization ability depends on the quality of the training data and the algorithm designer’s wisdom in choosing a particular operator form, i.e. how well it matches the function at hand. Scientific machine learning is a class of methods that constrain the neural network operator by forcing its output to match time-series data from a partially known dynamical model, e.g. an ordinary or partial vector differential equation. In this talk, we make the case for regularizing optical image measurements using this approach. Applications are expected to be in processes with high-complexity constitutive relationships, such as pharmaceutical and cell manufacturing, plant biology, and ecology.

Phase recovery (PR) refers to calculating the phase of the light field from its intensity measurements. As exemplified from quantitative phase imaging and coherent diffraction imaging to adaptive optics, PR is essential for reconstructing the refractive index distribution or topography of an object and correcting the aberration of an imaging system. In recent years, deep learning (DL), often implemented through deep neural networks, has provided unprecedented support for computational imaging, leading to more efficient solutions for various PR problems. In this review, we first briefly introduce conventional methods for PR. Then, we review how DL provides support for PR from the following three stages, namely, pre-processing, in-processing, and post-processing. We also review how DL is used in phase image processing. Finally, we summarize the work in DL for PR and provide an outlook on how to better use DL to improve the reliability and efficiency of PR. Furthermore, we present a live-updating resource ( https://github.com/kqwang/phase-recovery ) for readers to learn more about PR.

We utilized the pupil engineering method to enhance contrast of the sidelobe region in speckle correlations. Thus, required data collection for the speckle granularity probe is suppressed down to single frame.

We show that graph connectivity and the associated flow properties are mapped onto spatial and temporal fluctuations in the far field. The autocorrelation functions are processed by unsupervised and supervised algorithms to invert this mapping. Full-text article not available; see video presentation

We investigate the mapping between flows on complex graphs, such as retina vasculature, and spatial-temporal flucturations in the far field. Unsupervised and supervised learning algorithms can be used to invert these maps and obtain quantitative information about connectivity and blood flow. Full-text article not available; see video presentation

In freeze drying, thermal radiation has a significant effect on the drying process of vials located near the corner and edge of the trays, resulting in non-uniformity of the products. Understanding and being able to predict the impact of thermal radiation are therefore critical to accurate determination of the drying process endpoint given the variation in heat transfer of each vial. This article presents a new mechanistic model that describes complex thermal radiation during primary drying in conventional, microwave-assisted, and hybrid freeze drying. Modeling of thermal radiation employs the diffuse gray surface model and radiation network approach, which systematically and accurately incorporates simultaneous radiation exchange between every surface including the chamber wall and vials, allowing the framework to be seamlessly applied for analyzing various freeze-dryer designs. Model validation with data from the literature shows accurate prediction of the drying times for all vials, including inner, edge, and corner vials. The validated model is demonstrated for thermal radiation analysis and parametric studies to guide the design and optimization of freeze dryers.

We demonstrate a high-performance vendor-agnostic method for massively parallel solving of ensembles of ordinary differential equations (ODEs) and stochastic differential equations (SDEs) on GPUs. The method is integrated with a widely used differential equation solver library in a high-level language (Julia's DifferentialEquations.jl) and enables GPU acceleration without requiring code changes by the user. Our approach achieves state-of-the-art performance compared to hand-optimized CUDA-C++ kernels while performing 20--100$\times$ faster than the vectorizing map (vmap) approach implemented in JAX and PyTorch. Performance evaluation on NVIDIA, AMD, Intel, and Apple GPUs demonstrates performance portability and vendor-agnosticism. We show composability with MPI to enable distributed multi-GPU workflows. The implemented solvers are fully featured -- supporting event handling, automatic differentiation, and incorporation of datasets via the GPU's texture memory -- allowing scientists to take advantage of GPU acceleration on all major current architectures without changing their model code and without loss of performance. We distribute the software as an open-source library https://github.com/SciML/DiffEqGPU.jl