Closing the Gap Between Theory and Practice During Alternating Optimization for GANs.

Synthesizing high-quality and diverse samples is the main goal of generative models. Despite recent great progress in generative adversarial networks (GANs), mode collapse is still an open problem, and mitigating it will benefit the generator to better capture the target data distribution. This article rethinks alternating optimization in GANs, which is a classic approach to training GANs in practice. We find that the theory presented in the original GANs does not accommodate this practical solution. Under the alternating optimization manner, the vanilla loss function provides an inappropriate objective for the generator. This objective forces the generator to produce the output with the highest discriminative probability of the discriminator, which leads to mode collapse in GANs. To address this problem, we introduce a novel loss function for the generator to adapt to the alternating optimization nature. When updating the generator by the proposed loss function, the reverse Kullback-Leibler divergence between the model distribution and the target distribution is theoretically optimized, which encourages the model to learn the target distribution. The results of extensive experiments demonstrate that our approach can consistently boost model performance on various datasets and network structures.

Deep Spatial Transformation for Pose-Guided Person Image Generation and Animation.

Pose-guided person image generation and animation aim to transform a source person image to target poses. These tasks require spatial manipulation of source data. However, Convolutional Neural Networks are limited by the lack of ability to spatially transform the inputs. In this paper, we propose a differentiable global-flow local-attention framework to reassemble the inputs at the feature level. This framework first estimates global flow fields between sources and targets. Then, corresponding local source feature patches are sampled with content-aware local attention coefficients. We show that our framework can spatially transform the inputs in an efficient manner. Meanwhile, we further model the temporal consistency for the person image animation task to generate coherent videos. The experiment results of both image generation and animation tasks demonstrate the superiority of our model. Besides, additional results of novel view synthesis and face image animation show that our model is applicable to other tasks requiring spatial transformation. The source code of our project is available at https://github.com/RenYurui/Global-Flow-Local-Attention.

Distinct biological effects of different nanoparticles commonly used in cosmetics and medicine coatings.

BACKGROUND: Metal oxides in nanoparticle form such as zinc oxide and titanium dioxide now appear on the ingredient lists of household products as common and diverse as cosmetics, sunscreens, toothpaste, and medicine. Previous studies of zinc oxide and titanium dioxide in non-nanoparticle format using animals have found few adverse effects. This has led the FDA to classify zinc oxide as GRAS (generally recognized as safe) for use as a food additive. However, there is no regulation specific for the use of these chemicals in nanoparticle format. Recent studies, however, have begun to raise concerns over the pervasive use of these compounds in nanoparticle forms. Unfortunately, there is a lack of easily-adaptable screening methods that would allow for the detection of their biological effects. RESULTS: We adapted two image-based assays, a fluorescence resonance energy transfer-based caspase activation assay and a green fluorescent protein coupled-LC3 assay, to test for the biological effects of different nanoparticles in a high-throughput format. We show that zinc oxide nanoparticles are cytotoxic. We also show that titanium dioxide nanoparticles are highly effective in inducing autophagy, a cellular disposal mechanism that is often activated when the cell is under stress. CONCLUSION: We suggest that these image-based assays provide a method of screening for the biological effects of similar compounds that is both efficient and sensitive as well as do not involve the use of animals.

Mixing solutions in inkjet formed vesicles.

Controlling the contents of liposomes and vesicles is essential for their use in medicine, biotechnology, and basic research. Cargos such as proteins, DNA, and RNA are of growing interest for therapeutic applications as well as for fundamental studies of cellular organization and function, but controlled encapsulation and mixing of biomolecules within vesicles has been a challenge. Recently, microfluidic encapsulation has been shown to efficiently load arbitrary solutions of biomolecules into unilamellar vesicles. This method utilizes a piezoelectrically driven liquid jet to deform a planar bilayer and form a vesicle, with the fluid vortex formed by the jet mixing the solution in the jet with the surrounding solution. Here, we describe the equipment and protocol used for loading mixtures within unilamellar vesicles by microfluidic encapsulation, and we measure the encapsulated fraction to be 79+/-5% using a falling vesicle technique. Additionally, we find that the presence of a continuous flow from the nozzle and changes in actuation voltage polarity do not significantly affect the encapsulated fraction. These results help to guide current applications and future development of this microfluidic encapsulation technique for forming and loading unilamellar vesicles.

Inkjet formation of unilamellar lipid vesicles for cell-like encapsulation.

Encapsulation of macromolecules within lipid vesicles has the potential to drive biological discovery and enable development of novel, cell-like therapeutics and sensors. However, rapid and reliable production of large numbers of unilamellar vesicles loaded with unrestricted and precisely-controlled contents requires new technologies that overcome size, uniformity, and throughput limitations of existing approaches. Here we present a high-throughput microfluidic method for vesicle formation and encapsulation using an inkjet printer at rates up to 200 Hz. We show how multiple high-frequency pulses of the inkjet's piezoelectric actuator create a microfluidic jet that deforms a bilayer lipid membrane, controlling formation of individual vesicles. Variations in pulse number, pulse voltage, and solution viscosity are used to control the vesicle size. As a first step toward cell-like reconstitution using this method, we encapsulate the cytoskeletal protein actin and use co-encapsulated microspheres to track its polymerization into a densely entangled cytoskeletal network upon vesicle formation.

Dynamic control of needle-free jet injection.

Many modern pharmaceutical therapies such as vaccines and macromolecular drugs benefit from transdermal delivery. Conventional transdermal drug delivery via hypodermic needles causes pain, non-compliance, and potential contamination. Alternative transdermal strategies that deliver drugs in a quick, reliable, painless, and inexpensive way are needed. Jet injectors, which deliver drugs through the skin using a high-speed stream of liquid propelled by compressed springs or gasses, provide a needle-free method of trandermal drug delivery. However, poor reliability as well as painful bruising and bleeding characterize these devices, due in part to the high and constant jet velocity with which drugs are delivered. Toward improved reliability and reduced pain, we have developed a jet injector capable of dynamic control of jet velocity during a single injection pulse. Using this device, we demonstrate that temporal control of jet velocity leads to independent control of penetration depth, by adjusting time at high velocity, and delivered dose, by adjusting time at low velocity, in model materials. This dynamic control of jet velocity creates the potential for better control of needle-free injections, as demonstrated through injection studies on whole ex vivo human skin samples.

Unilamellar vesicle formation and encapsulation by microfluidic jetting.

Compartmentalization of biomolecules within lipid membranes is a fundamental requirement of living systems and an essential feature of many pharmaceutical therapies. However, applications of membrane-enclosed solutions of proteins, DNA, and other biologically active compounds have been limited by the difficulty of forming unilamellar vesicles with controlled contents in a repeatable manner. Here, we demonstrate a method for simultaneously creating and loading giant unilamellar vesicles (GUVs) using a pulsed microfluidic jet. Akin to blowing a bubble, the microfluidic jet deforms a planar lipid bilayer into a vesicle that is filled with solution from the jet and separates from the planar bilayer. In contrast with existing techniques, our method rapidly generates multiple monodisperse, unilamellar vesicles containing solutions of unrestricted composition and molecular weight. Using the microfluidic jetting technique, we demonstrate repeatable encapsulation of 500-nm particles into GUVs and show that functional pore proteins can be incorporated into the vesicle membrane to mediate transport. The ability of microfluidic jetting to controllably encapsulate solutions inside of GUVs creates new opportunities for the study and use of compartmentalized biomolecular systems in science, industry, and medicine.

Piezoelectric control of needle-free transdermal drug delivery.

Transdermal drug delivery occurs primarily through hypodermic needle injections, which cause pain, require a trained administrator, and may contribute to the spread of disease. With the growing number of pharmaceutical therapies requiring transdermal delivery, an effective, safe, and simple needle-free alternative is needed. We present and characterize a needle-free jet injector that employs a piezoelectric actuator to accelerate a micron-scale stream of fluid (40-130 microm diameter) to velocities sufficient for skin penetration and drug delivery (50-160 m/s). Existing jet injectors, powered by compressed springs and gases, are not widely used due to painful injections and poor reliability in skin penetration depth and dose. In contrast, our device offers electronic control of the actuator expansion rate, resulting in direct control of jet velocity and thus the potential for more precise injections. We apply a simple fluid-dynamic model to predict the device response to actuator expansion. Further, we demonstrate that injection parameters including expelled volume, jet pressure, and penetration depth in soft materials vary with actuator expansion rate, but are highly coupled. Finally, we discuss how electronically-controlled jet injectors may enable the decoupling of injection parameters such as penetration depth and dose, improving the reliability of needle-free transdermal drug delivery.