Search

Neural-based multi-task learning (MTL) has been successfully applied to many recommendation applications. However, these MTL models (e.g., MMoE, PLE) did not consider feature interaction during the optimization, which is crucial for capturing complex high-order features and has been widely used in ranking models for real-world recommender systems. Moreover, through feature importance analysis across various tasks in MTL, we have observed an interesting divergence phenomenon that the same feature can have significantly different importance across different tasks in MTL. To address these issues, we propose Deep Multiple Task-specific Feature Interactions Network (DTN) with a novel model structure design. DTN introduces multiple diversified task-specific feature interaction methods and task-sensitive network in MTL networks, enabling the model to learn task-specific diversified feature interaction representations, which improves the efficiency of joint representation learning in a general setup. We applied DTN to our company's real-world E-commerce recommendation dataset, which consisted of over 6.3 billion samples, the results demonstrated that DTN significantly outperformed state-of-the-art MTL models. Moreover, during online evaluation of DTN in a large-scale E-commerce recommender system, we observed a 3.28% in clicks, a 3.10% increase in orders and a 2.70% increase in GMV (Gross Merchandise Value) compared to the state-of-the-art MTL models. Finally, extensive offline experiments conducted on public benchmark datasets demonstrate that DTN can be applied to various scenarios beyond recommendations, enhancing the performance of ranking models.

Nonuniformly sampled signals are prevalent in real-world applications, but their power spectra estimation, usually from a finite number of samples of a single realization, presents a significant challenge. The optimal solution, which uses Bronez Generalized Prolate Spheroidal Sequence (GPSS), is computationally demanding and often impractical for large datasets. This paper describes a fast nonparametric method, the MultiTaper NonUniform Fast Fourier Transform (MTNUFFT), capable of estimating power spectra with lower computational burden. The method first derives a set of optimal tapers through cubic spline interpolation on a nominal analysis band. These tapers are subsequently shifted to other analysis bands using the NonUniform FFT (NUFFT). The estimated spectral power within the band is the average power at the outputs of the taper set. This algorithm eliminates the needs for time-consuming computation to solve the Generalized Eigenvalue Problem (GEP), thus reducing the computational load from $O(N^4)$ to $O(N \log N + N \log(1/\epsilon))$, which is comparable with the NUFFT. The statistical properties of the estimator are assessed using Bronez GPSS theory, revealing that the bias of estimates and variance bound of the MTNUFFT estimator are identical to those of the optimal estimator. Furthermore, the degradation of bias bound may serve as a measure of the deviation from optimality. The performance of the estimator is evaluated using both simulation and real-world data, demonstrating its practical applicability. The code of the proposed fast algorithm is available on GitHub (https://github.com/jiecui/mtnufft)., Comment: 22 pages, 6 figures and 1 table

The growing demand for real-time ultraviolet (UV) monitoring calls for a simple, rapid, and low-cost strategy to prepare UV photodetectors. We prepare a wearable real-time UV photodetector based on hydrogenated titanium dioxide film synthesized by radio frequency atmospheric pressure plasma. The conductivity of our hydrogenated titanium dioxide is improved to 10.2 S cm$^{-1}$, 9 orders of magnitude higher than that of pristine titanium dioxide after 10 min plasma treatment. Plasma hydrogenation disrupts the surface crystal structure, introducing oxygen vacancies (OVs) that create self-doped titanium(III) and titanium (II) species. First-principles calculations indicate the OVs raise the Fermi level of TiO$_2$ and distort the lattice nearby. Our optimized film has a distinctive periodic switching characteristic under intermittent illumination and good responsivity from 280 - 400 nm, peaking at 632.35 mA W^{-1}$ at 365 nm. The fabricated wearable sensor based on the optimized film effectively performs the monitoring of the daily variation of ambient UV intensity in three typical weather types and transferring its data to a smartphone via Wi-Fi.

Understanding the dynamics of the magnetic skyrmion, a particle-like topologically stable spin texture, and its response dynamics to external fields are indis-pensable for the applications in spintronic devices. In this letter, the Lorentz transmis-sion electron microscopy (LTEM) was used to investigate the spin chirality of the mag-netic skyrmion bubbles (SKBs) in the centrosymmetric magnet MnNiGa at room tem-perature. The reversal of SKBs excited by the in-plane magnetic field has been revealed. Moreover, the collective behavior of interacting spin chirality can be manipulated by reversing the directions of the magnetic fields on a wedge-shaped thin plate. The dy-namic behavior of the bubbles at different position of the thin plate has been explored with the micromagnetic simulation, indicating a non-uniform and nontrivial dynamic magnetization on the surfaces and center of the thin plate during the spin chirality re-versal. The results suggest that the controllable symmetry breaking of the SKBs arising from thickness variation provides an ability to manipulate the collective behavior of the spin chirality with small external fields, leading to a promising application in nonvola-tile spintronic devices for magnetic skyrmions.

Transport of intensity equation (TIE) has been applied to process the simulated and experimental images of the magnetic hard nanobubbles, which were acquired in the Lorentz transmission electron microscope (LTEM). Systematic studies demonstrated that the processing parameter in TIE can modulate the features of the retrieved magnetization and induce the bi-spiral structures which may be identified as the bi-skyrmions.

Spectrum auction is an effective approach to improving spectrum utilization, by leasing idle spectrum from primary users to secondary users. Recently, a few differentially private spectrum auction mechanisms have been proposed, but, as far as we know, none of them addressed the differential privacy in the setting of double spectrum auctions. In this paper, we combine the concept of differential privacy with double spectrum auction design, and present a Differentially private Double spectrum auction mechanism with approximate Social welfare Maximization (DDSM). Specifically, we design the mechanism by employing the exponential mechanism to select clearing prices for the double spectrum auction with probabilities exponentially proportional to the related social welfare values, and then improve the mechanism in several aspects like the designs of the auction algorithm, the utility function and the buyer grouping algorithm. Through theoretical analysis, we prove that DDSM achieves differential privacy, approximate truthfulness, approximate social welfare maximization. Extensive experimental evaluations show that DDSM achieves a good performance in term of social welfare., Comment: 12 pages, 7figures

Truthful spectrum auction is believed to be an effective method for spectrum redistribution. However, privacy concerns have largely hampered the practical applications of truthful spectrum auctions. In this paper, to make the applications of double spectrum auctions practical, we present a privacy-preserving and socially efficient double spectrum auction design, SDSA. Specifically, by combining three security techniques: homomorphic encryption, secret sharing and garbled circuits, we design a secure two-party protocol computing a socially efficient double spectrum auction, TDSA, without leaking any information about sellers' requests or buyers' bids beyond the auction outcome. We give the formal security definition in our context, and theoretically prove the security that our design achieves. Experimental results show that our design is also efficient in performance, even for large-scale double spectrum auctions., Comment: 11 pages, 6figures

A $(t,s,v)$-all-or-nothing transform (AONT) is a bijective mapping defined on $s$-tuples over an alphabet of size $v$, which satisfies that if any $s-t$ of the $s$ outputs are given, then the values of any $t$ inputs are completely undetermined. When $t$ and $v$ are fixed, to determine the maximum integer $s$ such that a $(t,s,v)$-AONT exists is the main research objective. In this paper, we solve three open problems proposed in [IEEE Trans. Inform. Theory 64 (2018), 3136-3143.] and show that there do exist linear $(2,p,p)$-AONTs. Then for the size of the alphabet being a prime power, we give the first infinite class of linear AONTs which is better than the linear AONTs defined by Cauchy matrices. Besides, we also present a recursive construction for general AONTs and a new relationship between AONTs and orthogonal arrays., Comment: arXiv admin note: text overlap with arXiv:1702.06612 by other authors

Chirping phenomena, in which the instantaneous frequencies of a signal change with time, are abundant in signals related to biological systems. Biosignals are non-stationary in nature and the time-frequency analysis is a viable tool to analyze them. It is well understood that Gaussian chirplet function is critical in describing chirp signals. Despite the theory of adaptive chirplet transform (ACT) has been established for more than two decades and is well accepted in the community of signal processing, application of ACT to bio-/biomedical signal analysis is still quite limited, probably because that the power of ACT, as an emerging tool for biosignal analysis, has not yet been fully appreciated by the researchers in the field of biomedical engineering. In this paper, we describe a novel ACT algorithm based on the "coarse-refinement" scheme. Namely, the initial estimate of a chirplet is implemented with the matching-pursuit (MP) algorithm and subsequently it is refined using the expectation-maximization (EM) algorithm, which we coin as MPEM algorithm. We emphasize the robustness enhancement of the algorithm in face of noise, which is important to biosignal analysis, as they are usually embedded in strong background noise. We then demonstrate the capability of our algorithm by applying it to the analysis of representative biosignals, including visual evoked potentials (bioelectrical signals), audible heart sounds and bat ultrasonic echolocation signals (bioacoustic signals), and human speech. The results show that the MPEM algorithm provides more compact representation of signals under investigation and clearer visualization of their time-frequency structures, indicating considerable promise of ACT in biosignal analysis. The MATLAB code repository is hosted on GitHub for free download (https://github.com/jiecui/mpact)., Comment: 27 pages, 8 figures

We explore the performance of a statistical learning technique based on Gaussian Process (GP) regression as an efficient non-parametric method for constructing multi-dimensional potential energy surfaces (PES) for polyatomic molecules. Using an example of the molecule N$_4$, we show that a realistic GP model of the six-dimensional PES can be constructed with only 240 potential energy points. We construct a series of the GP models and illustrate the convergence of the accuracy of the resulting surfaces as a function of the number of ${\it ab \ initio}$ points. We show that the GP model based on $\sim 1500$ potential energy points achieves the same level of accuracy as the conventional regression fits based on 16,421 points. The GP model of the PES requires no fitting of ${\it ab \ initio}$ data with analytical functions and can be readily extended to surfaces of higher dimensions., Comment: 2 figures

We consider a problem of extrapolating the collision properties of a large polyatomic molecule A-H to make predictions of the dynamical properties for another molecule related to A-H by the substitution of the H atom with a small molecular group X, without explicitly computing the potential energy surface for A-X. We assume that the effect of the $-$H $\rightarrow$ $-$X substitution is embodied in a multidimensional function with unknown parameters characterizing the change of the potential energy surface. We propose to apply the Gaussian Process model to determine the dependence of the dynamical observables on the unknown parameters. This can be used to produce an interval of the observable values that corresponds to physical variations of the potential parameters. We show that the Gaussian Process model combined with classical trajectory calculations can be used to obtain the dependence of the cross sections for collisions of C$_6$H$_5$CN with He on the unknown parameters describing the interaction of the He atom with the CN fragment of the molecule. The unknown parameters are then varied within physically reasonable ranges to produce a prediction uncertainty of the cross sections. The results are normalized to the cross sections for He - C$_6$H$_6$ collisions obtained from quantum scattering calculations in order to provide a prediction interval of the thermally averaged cross sections for collisions of C$_6$H$_5$CN with He., Comment: 7 figures

We show that a Gaussian Process model can be combined with a small number (of order 100) of scattering calculations to provide a multi-dimensional dependence of scattering observables on the experimentally controllable parameters such as the collision energy or temperature) as well as the potential energy surface (PES) parameters. For the case of Ar - C$_6$H$_6$ collisions, we show that 200 classical trajectory calculations are sufficient to provide a 10-dimensional hypersurface, giving the dependence of the collision lifetimes on the collision energy, internal temperature and 8 PES parameters. This can be used for solving the inverse scattering problem, the efficient calculation of thermally averaged observables, for reducing the error of the molecular dynamics calculations by averaging over the PES variations, and the analysis of the sensitivity of the observables to individual parameters determining the PES.Trained by a combination of classical and quantum calculations, the model provides an accurate description of the quantum scattering cross sections, even near scattering resonances., Comment: 3 figures

We calculate the cross sections for elastic scattering and Zeeman relaxation in binary collisions of molecules in the ro-vibrational ground state of a $^2\Sigma$ electronic state and the Zeeman state with the electron spin projection $M_S=1/2$ on the magnetic field axis. This is the lowest-energy state of $^2\Sigma$ molecules confined in a magnetic trap. The results are averaged over calculations with multiple molecule - molecule interaction potentials, which yields the expectation intervals for the cross sections and the elastic-to-inelastic cross section ratios. We find that the elastic-to-inelastic cross section ratios under conditions corresponding to trapped molecular ensembles at $T \sim 10^{-3}$ K exceed 100 for the majority of $^2\Sigma$ molecules. The range of $^2\Sigma$ molecules expected to be collisionally unstable in magnetic traps at $T < 10^{-3}$ K is limited to molecules with the spin-rotation interaction constant $\gamma_{\rm SR} > 0.5$ cm$^{-1}$ and the rotational constant $B_e < 4$ cm$^{-1}$.

We present expressions demonstrating that collisional decoherence of ultracold atoms or molecules in a coherent superposition of non-degenerate quantum states is suppressed when both the real and imaginary parts of the scattering lengths for the states in the coherent superposition are equal. We show that the rate of collisional decoherence can be enhanced or suppressed by varying an external magnetic field near a Feshbach resonance. For some resonances, the suppression is very dramatic. We propose a method for measuring the scattering length of ultracold particles in excited quantum states exhibiting Feshbach resonances.