Search

Generative adversarial nets (GAN) has been successfully introduced for generating text to alleviate the exposure bias. However, discriminators in these models only evaluate the entire sequence, which causes feedback sparsity and mode collapse. To tackle these problems, we propose a novel mechanism. It first segments the entire sequence into several sub-sequences. Then these sub-sequences, together with the entire sequence, are evaluated individually by the discriminator. At last these feedback signals are all used to guide the learning of GAN. This mechanism learns the generation of both the entire sequence and the sub-sequences simultaneously. Learning to generate sub-sequences is easy and is helpful in generating an entire sequence. It is easy to improve the existing GAN-based models with this mechanism. We rebuild three previous well-designed models with our mechanism, and the experimental results on benchmark data show these models are improved significantly, the best one outperforms the state-of-the-art model.\footnote[1]{All code and data are available at https://github.com/liyzcj/seggan.git

Based on the coupled-dipole analysis and finite-difference time-domain simulation, we have investigated the surface plasmon propagation in one-dimensional metallic nanoparticle (NP) chains. Our systematic studies reveal that the interplay between the localized plasmon excitation and the lattice collective behavior leads to two phases (I and II) of different electromagnetic (EM) field transport properties. In phase I, the EM field decays follow the power-law. In phase II, the EM field shows the exponential decay in the short distance regime and the power-law decay in the long distance regime. Moreover, universal power-law exponents have been found in the long propagation distance. The two different EM field propagation behaviors can be transformed to each other by tuning the parameters of the excitation fields and/or those of the NP chains. The EM field transport mechanisms we have found are very useful in the design of plasmonic waveguide with both strong field confinement and efficient field/energy transfer, which has important applications in integrated nanophotonic circuits., Comment: 12 pages, 11 figures

We investigate the strong coupling phenomenon in Kretschmann configuration involving molecular J-aggregates by using surface plasmon polaritons (SPPs) as a light source. SPPs are injected in molecular J-aggregates and propagate along the interface between molecular J-aggregates and the metal film. Both the classical method and Finite difference time domain method are used to show the radiative characteristics of light converted from SPPs in our proposed structure, with the results from the two methods being consistent with each other. Our results show that the radiative angle versus the wavelength shows a Fano line shape and the conversion efficiency of SPPs into light exhibits the strong coupling phenomenon in form of Rabi splitting. Our structure is quite sensitive to the wavelength, which has potential applications in sensing, lasing, optical switching, and nonlinear slow-light devices.

Hotspots with strong near fields due to localized surface plasmons (LSPs) in metallic nanostructures have various applications, such as surface enhanced Raman scattering (SERS). The long range Coulomb coupling between LSPs in periodic metallic nanostructures may lead to interesting collective effects. In this paper, we investigate the combination effects of the local field enhancement and collective plasmon resonances in one dimensional metallic nanoparticle dimer chains. It is found that the strong near field in the gap and the far field interactions among the metallic nanoparticles lead to anomalous collective hotspots with dual enhancement of the electromagnetic field. In particular, the interference between the incident field and the induced internal field leads to Fano-type effect with Wood anomaly related destructive interference and the strong resonance with an extremely narrow width. Our systematic study shows that the correlation between the local structure and the global structure has important impact on the collective spots, which leads to an optimal orientation of the dimer (about 60{\deg} with respect to the chain direction) for the largest gap field enhancement with the incident field polarization parallel to the long axis of the dimer., Comment: 14 pages, 7 figures

Ongoing research in NASA Glenn Research Center's Structural Mechanics and Dynamics Branch to develop smart materials technologies for aeropropulsion structural components has resulted in the design of the prototype morphing fan nozzle shown in the photograph. This prototype exploits the potential of smart materials to significantly improve the performance of existing aircraft engines by introducing new inherent capabilities for shape control, vibration damping, noise reduction, health monitoring, and flow manipulation. The novel design employs two different smart materials, a shape-memory alloy and magnetorheological fluids, to reduce the nozzle area by up to 30 percent. The prototype of the variable-area fan nozzle implements an overlapping spring leaf assembly to simplify the initial design and to provide ease of structural control. A single bundle of shape memory alloy wire actuators is used to reduce the nozzle geometry. The nozzle is subsequently held in the reduced-area configuration by using magnetorheological fluid brakes. This prototype uses the inherent advantages of shape memory alloys in providing large induced strains and of magnetorheological fluids in generating large resistive forces. In addition, the spring leaf design also functions as a return spring, once the magnetorheological fluid brakes are released, to help force the shape memory alloy wires to return to their original position. A computerized real-time control system uses the derivative-gain and proportional-gain algorithms to operate the system. This design represents a novel approach to the active control of high-bypass-ratio turbofan engines. Researchers have estimated that such engines will reduce thrust specific fuel consumption by 9 percent over that of fixed-geometry fan nozzles. This research was conducted under a cooperative agreement (NCC3-839) at the University of Akron.