Memory Analysis for Memristors and Memristive Recurrent Neural Networks

Traditional recurrent neural networks are composed of capacitors, inductors, resistors, and operational amplifiers.Memristive neural networks are constructed by replacing resistors with memristors. This paper focuses on the memory analysis,i.e. the initial value computation, of memristors. Firstly, we present the memory analysis for a single memristor based on memristors’ mathematical models with linear and nonlinear drift.Secondly, we present the memory analysis for two memristors in series and parallel. Thirdly, we point out the difference between traditional neural networks and those that are memristive. Based on the current and voltage relationship of memristors, we use mathematical analysis and SPICE simulations to demonstrate the validity of our methods.

Cooperative Loss Recovery for Reliable Multicast in Ad Hoc Networks

Providing reliable multicast service is very challenging in Ad Hoc networks. In this paper, we propose an efficient loss recovery scheme for reliable multicast (CoreRM). Our basic idea is to apply the notion of cooperative communications to support local loss recovery in multicast. A receiver node experiencing a packet loss tries to recover the lost packet through progressively cooperating with neighboring nodes, upstream nodes or even source node. In order to reduce recovery latency and retransmission overhead, CoreRM caches not only data packets but also the path which could be used for future possible use to expedite the loss recovery process. Both analytical and simulation results reveal that CoreRM significantly improves the reliable multicast performance in terms of delivery ratio, throughput and recovery latency compared with UDP and PGM.

Droplet Characterization Based on the Simulated Secondary Rainbows

The droplet size, size distribution, refractive index, and temperature can be measured simultaneously by the rainbow technique. In the present work, the rainbow scattering diagram for a spherical droplet in the secondary rainbow region is simulated by the use of the generalized Lorenz-Mie theory. For achieving high spatial resolution in denser droplet sprays, a focused Gaussian beam is used. For droplet characterization, different inversion algorithms are investigated, which includes trough-trough (θmin1 and θmin2) method and inflection-inflection (θinf1 and θinf2) method. For the trough-trough algorithm, the absolute error of the refractive index is between −6.4 × 10−4 and 1.7 × 10−4, and the error of the droplet radius is only between −0.55% and 1.77%. For the inflection-inflection algorithm, the maximum absolute error of the inverted refractive index is less than −1.1 × 10−3. The error of the droplet radius is between −0.75% and 5.67%.

Non-invasive photoacoustic computed tomography of rat heart anatomy and function

Complementary to mainstream cardiac imaging modalities for preclinical research,photoacoustic computed tomography(PACT)can provide functional optical contrast with high imaging speed and resolution.However,PACT has not been demonstrated to reveal the dynamics of whole cardiac anatomy or vascular system without surgical procedure(thoracotomy)for tissue penetration.Here,we achieved non-invasive imaging of rat hearts using the recently developed three-dimensional PACT(3D-PACT)platform,demonstrating the regulated illumination and detection schemes to reduce the effects of optical attenuation and acoustic distortion through the chest wall;thereby,enabling unimpeded visualization of the cardiac anatomy and intracardiac hemodynamics following rapidly scanning the heart within 10 s.We further applied 3D-PACT to reveal distinct cardiac structural and functional changes among the healthy,hypertensive,and obese rats,with optical contrast to uncover differences in cardiac chamber size,wall thickness,and hemodynamics.Accordingly,3D-PACT provides high imaging speed and nonionizing penetration to capture the whole heart for diagnosing the animal models,holding promises for clinical translation to cardiac imaging of human neonates.

Single-pulse real-time billion-frames-per-second planar imaging of ultrafast nanoparticle-laser dynamics and temperature in flames

Unburnt hydrocarbon flames produce soot,which is the second biggest contributor to global warming and harmful to human health.The state-of-the-art high-speed imaging techniques,developed to study non-repeatable turbulent flames,are limited to million-frames-per-second imaging rates,falling short in capturing the dynamics of critical species.Unfortunately,these techniques do not provide a complete picture of flame-laser interactions,important for understanding soot formation.Furthermore,thermal effects induced by multiple consecutive pulses modify the optical properties of soot nanoparticles,thus making single-pulse imaging essential.Here,we report single-shot laser-sheet compressed ultrafast photography(LS-CUP)for billion-frames-per-second planar imaging of flame-laser dynamics.We observed laser-induced incandescence,elastic light scattering,and fluorescence of soot precursors-polycyclic aromatic hydrocarbons(PAHs)in real-time using a single nanosecond laser pulse.The spatiotemporal maps of the PAHs emission,soot temperature,primary nanoparticle size,soot aggregate size,and the number of monomers,present strong experimental evidence in support of the theory and modeling of soot inception and growth mechanism in flames.LS-CUP represents a generic and indispensable tool that combines a portfolio of ultrafast combustion diagnostic techniques,covering the entire lifecycle of soot nanoparticles,for probing extremely short-lived(picoseconds to nanoseconds)species in the spatiotemporal domain in non-repeatable turbulent environments.Finally,LS-CUP’s unparalleled capability of ultrafast wide-field temperature imaging in real-time is envisioned to unravel mysteries in modern physics such as hot plasma, sonoluminescence, and nuclear fusion.

Deep learning-enhanced microscopy with extended depth-of-field

A computational imaging platform utilizing a physics-incorporated,deep-learned design of binary phase filter and a jointly optimized deconvolution neural network has been reported,achieving high-resolution,high-contrast imaging over extended depth ranges without the need for serial refocusing.