Search

Embedding Based Retrieval (EBR) is a crucial component of the retrieval stage in (Ads) Recommendation System that utilizes Two Tower or Siamese Networks to learn embeddings for both users and items (ads). It then employs an Approximate Nearest Neighbor Search (ANN) to efficiently retrieve the most relevant ads for a specific user. Despite the recent rise to popularity in the industry, they have a couple of limitations. Firstly, Two Tower model architecture uses a single dot product interaction which despite their efficiency fail to capture the data distribution in practice. Secondly, the centroid representation and cluster assignment, which are components of ANN, occur after the training process has been completed. As a result, they do not take into account the optimization criteria used for retrieval model. In this paper, we present Hierarchical Structured Neural Network (HSNN), a deployed jointly optimized hierarchical clustering and neural network model that can take advantage of sophisticated interactions and model architectures that are more common in the ranking stages while maintaining a sub-linear inference cost. We achieve 6.5% improvement in offline evaluation and also demonstrate 1.22% online gains through A/B experiments. HSNN has been successfully deployed into the Ads Recommendation system and is currently handling major portion of the traffic. The paper shares our experience in developing this system, dealing with challenges like freshness, volatility, cold start recommendations, cluster collapse and lessons deploying the model in a large scale retrieval production system., Comment: 9 pages

Recently, an extremely-air-stable one-dimensional 7-9-7-AGNR was successfully fabricated. To further reveal its potential application in sub-5-nm field-effect transistors (FETs), there is an urgent need to develop integrated circuits. Here, we report first-principles quantum-transport simulations on the performance limits of n- and p-type sub-5-nm one-dimensional 7-9-7-AGNR FET. We find that the on-state current (Ion) in 7-9-7-AGNR FET can be effectively manipulated by the length of the gate and underlap. Particularly, the optimized Ion in n-type (p-type) device can reach up to 2423 (4277) and 1988 (920) {\mu}A/{\mu}m for high-performance and low-power applications, respectively. The large Ion values are in the first class among the LD FETs, which can well satisfy the ITRS requirements. We also find that the 7-9-7-AGNR FET can have ultralow subthreshold swing below 60mV/dev, ultrashort delay time (<0.01 ps), and very small power-delay product (<0.01 fJ/{\mu}m). Our results show that the 7-9-7-AGNR based FETs have great potential applications in the high-speed and low-power consumption chips., Comment: 7 figures, 2 tables

Generative AI tools, such as ChatGPT and Midjourney, are transforming artistic creation as AI-art integration advances. However, Artificial Intelligence Generated Content (AIGC) tools face user experience challenges, necessitating a human-centric design approach. This paper offers a brief overview of research on explainable AI (XAI) and user experience, examining factors leading to suboptimal experiences with AIGC tools. Our proposed solution integrates interpretable AI methodologies into the input and adjustment feedback stages of AIGC products. We underscore XAI's potential to enhance the user experience for ordinary users and present a conceptual framework for improving AIGC user experience.

China adjusted the zero-COVID strategy in late 2022, triggering an unprecedented Omicron wave. We estimated the time-varying reproduction numbers of 32 provincial-level administrative divisions from December 2022 to January 2023. We found that the pooled estimate of initial reproduction numbers is 4.74 (95% CI: 4.41, 5.07).

Applying the system of linear partial differential equations derived from the Mellin-Barnes representation and the Miller transformation, we present the GKZ-system of the Feynman integral of the 2-loop self energy diagram with 4 propagators. The codimension of the derived GKZ-system equals the number of independent dimensionless ratios among the external momentum squared and virtual mass squared. In total 536 hypergeometric functions are obtained in the neighborhoods of the origin and infinity, in which 30 linearly independent hypergeometric functions whose convergent regions have nonempty intersection constitute a fundamental solution system in a proper subset of the whole parameter space., Comment: latex, 306 pages, including 1 figure + 254 pages of supplementary material, accepted for publication in the European Physical Journal C. arXiv admin note: text overlap with arXiv:2206.04224

The magnetic fields with the first- and second-order gradient are engineered in several mechanically controlled hybrid systems. The current-carrying nanowires with different geometries can induce a tunable magnetic field gradient because of their geometric symmetries, and therefore develop various couplings to nitrogen-vacancy (NV) centers. For instance, a straight nanowire can guarantee the Jaynes-Cummings (JC) spin-phonon interaction and may indicate a potential route towards the application on quantum measurement. Especially, two parallel straight nanowires can develop the coherent down-conversion spin-phonon interaction through a second-order gradient of the magnetic field, and it can induce a bundle emission of the antibunched phonon pairs via an entirely different magnetic mechanism. Maybe, this investigation is further believed to support NV's future applications in the area of quantum manipulation, quantum sensing, and precision measurement, etc., Comment: This manuscript need to be modified

The investigation on significantly enhancing the coupling to NV centers at single-quanta level is of great key point to further explore its application in quantum information processing (QIP). We here study a joint scheme to further enhance NV-phonon coherent coupling with two methods working together in a hybrid optomechanical systems. Both methods are mechanic-induced mode field coupling (MFC) leading to the modification of the spatial distribution of the optical field, and the mechanical parametric amplification (MPA) realized by modulating the mechanical spring constant in time, respectively. With the joint assistance of MFC and MPA, the coherent coupling between the NV spin and one supermode of the mechanical resonators (MRs) can be further enhanced significantly, with the rate ne^r, several potential applications on this proposal are also discussed in this work. For the ultimate target of enhancing the coupling to NV spin at single-quanta level, this attempt may provide a promising spin-phonon platform for implementing the more active control.

Sparse random linear network coding (SRLNC) used as a class of erasure codes to ensure the reliability of multicast communications has been widely investigated. However, an exact expression for the decoding success probability of SRLNC is still unknown, and existing expressions are either asymptotic or approximate. In this paper, we derive an exact expression for the decoding success probability of SRLNC. The key to achieving this is to propose a criterion that a vector is contained in a subspace. To obtain this criterion, we construct a basis of a subspace, with respect to this basis, the coordinates of a vector are known, based on a maximal linearly independent set of the columns of a matrix. The exactness and the computation of the derived expression are demonstrated by a simple example., Comment: arXiv admin note: text overlap with arXiv:2010.05555

Event cameras are a kind of bio-inspired sensors that generate data when the brightness changes, which are of low-latency and high dynamic range (HDR). However, due to the nature of the sparse event stream, event-based mapping can only obtain sparse or semi-dense edge 3D maps. By contrast, standard cameras provide complete frames. To leverage the complementarity of event-based and standard frame-based cameras, we propose a fusion strategy for dense mapping in this paper. We first generate an edge map from events, and then fill the map using frames to obtain the dense depth map. We propose "filling score" to evaluate the quality of filled results and show that our strategy can increase the number of existing semi-dense 3D map., Comment: 4 pages, 2 figures, 6 references

Reliable and low latency multicast communication is important for future vehicular communication. Sparse random linear network coding approach used to ensure the reliability of multicast communication has been widely investigated. A fundamental problem of such communication is to characterize the decoding success probability, which is given by the probability of a sparse random matrix over a finite field being full rank. However, the exact expression for the probability of a sparse random matrix being full rank is still unknown, and existing approximations are recursive or not consistently tight. In this paper, we provide a tight and closed-form approximation to the probability of a sparse random matrix being full rank, by presenting the explicit structure of the reduced row echelon form of a full rank matrix and using the product theorem. Simulation results show that our proposed approximation is of high accuracy regardless of the generation size, the number of coded packets, the field size and the sparsity, and tighter than the state-of-the-art approximations for a large range of parameters.

The knots in solar prominences are often observed to fall with nearly constant velocity, but the associated physical mechanism is currently not well understood. In this letter, we presented a prominence observed by New Vacuum Solar Telescope (NVST) in H-alpha wavelength. Knots that rose within the prominence appear to have been preferentially located at higher altitude, whereas those that fell were found throughout the entire prominence structure. The descending speed of the knots near the solar surface was higher than that far away from the solar surface. We noted that the knots near the solar surface may run along a set of coronal loops observed from the Atmospheric Imaging Assembly. Elsewhere, the majority of knots are interpreted to have descended across more horizontal magnetic field with a nearly constant speed. This lack of acceleration indicates that the liberated gravitational potential energy may not manifest as an increase in kinetic energy. Assuming instead that the descending knots were capable of exciting Alfven waves that could then dissipate within the local corona, the gravitational potential energy of the knots may have been converted into thermal energy. Assuming a perfectly elastic system, we therefore estimate that the gravitational energy loss rate of these observed knots amounts to 1/2000 of that required to heat the entire quiet-Sun, increasing to 1/320 when considering possibly further downward motions of the knots having disappeared in the H-alpha observations. This result suggests such a mechanism may contribute to the heating of the corona local to these prominences., Comment: 13 pages, 6 figures

In speech synthesis and speech enhancement systems, melspectrograms need to be precise in acoustic representations. However, the generated spectrograms are over-smooth, that could not produce high quality synthesized speech. Inspired by image-to-image translation, we address this problem by using a learning-based post filter combining Pix2PixHD and ResUnet to reconstruct the mel-spectrograms together with super-resolution. From the resulting super-resolution spectrogram networks, we can generate enhanced spectrograms to produce high quality synthesized speech. Our proposed model achieves improved mean opinion scores (MOS) of 3.71 and 4.01 over baseline results of 3.29 and 3.84, while using vocoder Griffin-Lim and WaveNet, respectively.

This paper proposes a novel model-free screening procedure for ultrahigh dimensional data analysis. By utilizing slicing technique which has been successfully ap- plied to continuous variables, we construct a new index called the fused mean-variance for feature screening. This method has the following merits: (i) it is model-free, i.e., without specifying regression form of predictors and response variable; (ii) it can be used to analyze various types of variables including discrete, categorical and continuous vari- ables; (iii) it still works well even when the covariates/random errors are heavy-tailed or the predictors are strongly dependent. Under some regularity conditions, we establish the sure screening and rank consistency. Simulation studies are conducted to assess the performance of the proposed approach. A real data is used to illustrate the proposed method.

Sn is a well-known classical superconductor on the border between type I and type II with critical temperature of 3.722K and critical field of 0.031T. We show by means of specific heat and electric magneto-transport data that its critical parameters can be dramatically increased if it is brought in the form of loosely bound bundles of thin nanowires. The specific heat displays a pronounced double phase transition at 3.7K and 5.5K, which we attribute to the inner 'bulk' contribution of the nanowires and to the surface contribution, respectively. The latter is visible only because of the large volume fraction of the surface layer in relation to their bulk inner volume. The upper transition coincides with the onset of the resistive transition, while zero resistance is gradually approached below the lower transition. The large coherence length of 230nm at 0K likely actuates a Josephson coupling between adjacent neighboring nanowires and thus suppresses the effect of 1D phase fluctuations along the nanowires, and stabilizes 3D phase coherence throughout the entire network with zero resistance. A magnetic field of more than 3T is required to restore the normal state, which means that the critical field is enhanced by about two orders of magnitude with respect to Sn in its bulk form.

The aim of this paper is to consider the linear ultraparabolic equation with bounded and VMO coefficients $a_{ij} (z)$. Assume that the operator $L_0$ obtained by freezing the coefficients $a_{ij}(z)$ at any point ${z_0} \in {\mathbb{R}^{N + 1}}$ is hypoelliptic. We first establish a Caccioppoli type inequality by choosing a cutoff function, a Sobolev type inequality by prosperities of the fundamental solution to $L_0$, and a Poincar\'{e} type inequality with a new cutoff function. Then $L^p$ estimate for weak solutions is derived by using the reverse H\"{o}lder inequality on homogeneous spaces. Finally, higher Morrey estimates for weak solutions to the above equation are shown by investigating a homogeneous ultraparabolic equation of variable coefficients with a nonhomogeneous boundary value condition, and a nonhomogeneous ultraparabolic equation of variable coefficients with homogeneous boundary value condition.

The aim of this paper is to establish regularity for weak solutions to the nondiagonal quasilinear degenerate elliptic systems related to H\"{o}rmander's vector fields, where the coefficients are bounded with vanishing mean oscillation. We first prove $L^p$($p \ge 2$) estimates for gradients of weak solutions by using a priori estimates and a known reverse H\"{o}lder inequality, and consider regularity to the corresponding nondiagonal homogeneous degenerate elliptic systems. Then we get higher Morrey and Campanato estimates for gradients of weak solutions to original systems and H\"{o}lder estimates for weak solutions.

Let $\mathrm{X}=(X_{1},...,X_{q})$ be a family of real smooth vector fields satisfying H\"{o}mander's condition. The purpose of this paper is to establish gradient estimates in generalized Morrey spaces for weak solutions of the divergence degenerate parabolic system related to $X$ :%\[u_{t}^{i}+X_{\alpha}^{\ast}(a_{ij}^{\alpha\beta}(z)X_{\beta}u^{j}%)=g_{i}+X_{\alpha}^{\ast}f_{i}^{\alpha}(z), \] where $\alpha,\beta=1,2,...,q,$ $i,j=1,2,...,N$, $X_{\alpha}^{\ast}$ is the transposed vector field of $X_{\alpha}$, $z=(t,x)\in{\mathbb{R}}^{n+1}$, and coefficients $a_{ij}^{\alpha\beta}(z)$ belong to the space $VMO$ induced by the vector fields $X_{1}, ...,X_{q}$., Comment: 21 pages