Search

Deep Neural Network (DNN) is a critical component of a wide range of applications. However, with the rapid growth of the training dataset and model size, communication becomes the bottleneck, resulting in low utilization of computing resources. To accelerate communication, recent works propose to aggregate gradients from multiple workers in the programmable switch to reduce the volume of exchanged data. Unfortunately, since using synchronization transmission to aggregate data, current in-network aggregation designs suffer from the straggler problem, which often occurs in shared clusters due to resource contention. To address this issue, we propose a straggler-aware aggregation transport protocol (SA-ATP), which enables the leading worker to leverage the spare computing and storage resources to help the straggling worker. We implement SA-ATP atop clusters using P4-programmable switches. The evaluation results show that SA-ATP reduces the iteration time by up to 57% and accelerates training by up to $1.8\times $ in real-world benchmark models.

Lithium niobate modulators have primarily been fabricated on chip-level or 4/6-inch wafers. Here, thin-film lithium niobate (TFLN) electro-optic Mach-Zehnder modulators (MZM) are demonstrated for the first time on an 8-inch silicon substrate. The fabricated LN modulator has an on-chip loss of less than 1 dB with a waveguide loss of lower than 0.5 dB/cm. The half-wave voltage- length product (V $\pi \cdot $ L) is 3.12 V $\cdot $ cm in the C-band at a 0.5 cm modulation length, and the corresponding 3-dB bandwidth of the electro-optic response is beyond 67 GHz. All specifications are state-of-the-art. These results provide the basis for the industrialization of the TFLN platform with better balance between performance and cost.

Metasurfaces, optics made from subwavelength-scale nanostructures, have been limited to millimeter-sizes by the scaling challenge of producing vast numbers of precisely engineered elements over a large area. In this study, we demonstrate an all-glass 100 mm diameter metasurface lens (metalens) comprising 18.7 billion nanostructures that operates in the visible spectrum with a fast f-number (f/1.5, NA = 0.32) using deep-ultraviolet (DUV) projection lithography. Our work overcomes the exposure area constraints of lithography tools and demonstrates that large metasurfaces are commercially feasible. Additionally, we investigate the impact of various fabrication errors on the imaging quality of the metalens, several of which are specific to such large area metasurfaces. We demonstrate direct astronomical imaging of the Sun, the Moon, and emission nebulae at visible wavelengths and validate the robustness of such metasurfaces under extreme environmental thermal swings for space applications.

• Inverse optimal method overcomes the difficulties of solving the nonlinear HJB equation. Not only are design procedure for 2-nd order systems given, but results for higher order systems are also generalised. • Integration of disturbance estimation in the virtual control design of the backstepping method, realizing the treatment of mismatched disturbances. • The composite inverse optimal control method gives a stabilizing control first with optimality and robustness, and then find the corresponding cost function. Controllers given in advance achieve well control performance with physical constraints, such as inpur saturation. • The proposed controller has a well understood and clear structure, which makes it easier for engineers to implement for practical systems. [Display omitted] The Hamilton-Jacobi-Bellman (HJB) equations in optimal control for interconnected systems are difficult to solve, and the presence of external disturbances makes the problem even more challenging. This paper proposes a decentralized control method using inverse optimal strategy for a class of interconnected systems with disturbances. The robustness of the system is maintained by using disturbance observers to estimate unknown disturbances. By using inverse optimal control to find the controller and corresponding cost function, we eliminate the need to solve the HJB equations. By Lyapunov theory, we prove the designed composite controller is of optimality and robustness. Furthermore, the results are extended to n − dimensional interconnected systems with rigorous proof. Finally, the effectiveness of the proposed method is demonstrated through two examples. [ABSTRACT FROM AUTHOR]

Identifying heavy flows is essential for network management. However, it is challenging to detect heavy flow quickly and accurately under the highly dynamic traffic and rapid growth of network capacity. Existing heavy flow detection schemes can make a trade-off in efficiency, accuracy and speed. However, these schemes still require memory large enough to obtain acceptable performance. To address this issue, we propose ChainSketch, which has the advantages of good memory efficiency, high accuracy and fast detection. Specifically, ChainSketch uses the selective replacement strategy to mitigate the over-estimation issue. Meanwhile, ChainSketch utilizes the hash chain and compact structure to improve memory efficiency. We implement the ChainSketch on OVS platform, P4-based testbed and large-scale simulations to process heavy hitter and heavy changer detection. The results of trace-driven tests show that, ChainSketch greatly improves the F1-score by up to $3.43\times $ compared with the state-of-the-art solutions especially for small memory.

In this talk, we will present a new platform for a virtual/augmented reality system based on a metalens and a fiber scanning display. The metalens was designed for achromatic focusing of blue, green, and red light. The fiber scanning display features combined desired merits of high resolution, high brightness, high dynamic range and wide color gamut.

This study aims to compare the dispersion of coughed droplets with different diameters under two heating strategies of hospital ward in winter. An experimentally verified computational fluid dynamics (CFD) method has been used to simulate the evaporation and diffusion of droplets with five different diameters in a two-bed ward with air conditioning strategy (MV) and floor radiant heating strategy (FH). Results show that in the boundary of the human body, indoor air flow, human expiratory airflow and thermal plume play an important role in the dispersion of coughed droplets. Small and medium-sized droplets suspension in the ward for a longer time, while large droplets deposit rapidly. The removal ratio of 50 μm droplets under MV strategies is the highest. The 100 μm droplets in the two heating strategies are completely removed within 30 s. There is an obvious difference of droplets deposition ratio on each surface in the ward.

Research on two-dimensional designer optical structures, or metasurfaces, has mainly focused on controlling the wavefronts of light propagating in free space. Here, we show that gradient metasurface structures consisting of phased arrays of plasmonic or dielectric nanoantennas can be used to control guided waves via strong optical scattering at subwavelength intervals. Based on this design principle, we experimentally demonstrate waveguide mode converters, polarization rotators and waveguide devices supporting asymmetric optical power transmission. We also demonstrate all-dielectric on-chip polarization rotators based on phased arrays of Mie resonators with negligible insertion losses. Our gradient metasurfaces can enable small-footprint, broadband and low-loss photonic integrated devices.