Search

Large language models, initially pre-trained with a limited context length, can better handle longer texts by continuing training on a corpus with extended contexts. However, obtaining effective long-context data is challenging due to the scarcity and uneven distribution of long documents across different domains. To address this issue, we propose a Query-centric data synthesis method, abbreviated as Quest. Quest is an interpretable method based on the observation that documents retrieved by similar queries are relevant but low-redundant, thus well-suited for synthesizing long-context data. The method is also scalable and capable of constructing large amounts of long-context data. Using Quest, we synthesize a long-context dataset up to 128k context length, significantly outperforming other data synthesis methods on multiple long-context benchmark datasets. In addition, we further verify that the Quest method is predictable through scaling law experiments, making it a reliable solution for advancing long-context models.

Receiving timely and personalized feedback is essential for second-language learners, especially when human instructors are unavailable. This study explores the effectiveness of Large Language Models (LLMs), including both proprietary and open-source models, for Automated Essay Scoring (AES). Through extensive experiments with public and private datasets, we find that while LLMs do not surpass conventional state-of-the-art (SOTA) grading models in performance, they exhibit notable consistency, generalizability, and explainability. We propose an open-source LLM-based AES system, inspired by the dual-process theory. Our system offers accurate grading and high-quality feedback, at least comparable to that of fine-tuned proprietary LLMs, in addition to its ability to alleviate misgrading. Furthermore, we conduct human-AI co-grading experiments with both novice and expert graders. We find that our system not only automates the grading process but also enhances the performance and efficiency of human graders, particularly for essays where the model has lower confidence. These results highlight the potential of LLMs to facilitate effective human-AI collaboration in the educational context, potentially transforming learning experiences through AI-generated feedback.

In this paper, we study the influence of the parameters for the accelerating Kerr-Newman black hole on the shadows and the constraints, extensively. We find that the rotating parameter $a$, the charge parameter $e$, and the inclination angle $\theta_0$ affect the shadow qualitatively similar to that of Kerr-Newman black holes. The result shows that the size of the shadow will scale down with the accelerating factor $A$. Besides, the factor $A$ also can affect the best viewing angles, which make the observations maximum deviate from $\theta_0=\frac{\pi}{2}$, and the degree of the deviations are less than $1\%$. Then, we assume the M87* as an accelerating Kerr-Newman black hole with the mass $M=6.5\times10^9M_\odot$ and the distance $r_0=16.8Mpc$. Combining the EHT observations, we find that neither the observations, circularity deviation $\Delta C$ or axial ratio $D_x$ can distinguish the accelerating black hole or not. However, the characteristic areal-radius of the shadow curve $R_a$ can give corresponding constraints on the parameters of the accelerating Kerr-Newman black hole. The results shows that the bigger accelerating factor $A$ is, the stronger constraints on the rotating parameter $a$ and charged parameter $e$. {The maximum range of the accelerating factor is $Ar_0\leq0.558$ for a accelerating Schwarzschild case with $(a/M=e/M=0)$, and for an extremely slow accelerating case $(Ar_0\leq0.01)$, the ranges of rotating parameter $a$ and charged parameter $e$ are $a/M\in(0,1)$ and $e/M\in(0,0.9)$., Comment: 9 pages, 16figures

In this paper, we present several explicit reconstructions for {the aether scalar tensor (AeST) theory} derived from the background of Friedmann-Lema$\hat{\text{\i}}$tre-Robertson-Walker cosmological evolution. It is shown that the Einstein-Hilbert Lagrangian with a positive cosmological constant is the only Lagrangian capable of accurately replicating the exact expansion history of the $\Lambda$ cold dark matter ($\Lambda$CDM) universe filled solely with dust-like matter. However, the $\Lambda$CDM-era can be produced within the framework of the AeST theory for some other fluids, including a perfect fluid with $p=-(1/3)\rho$, multifluids, and nonisentropic perfect fluids. Moreover, we demonstrate that the $\Lambda$CDM-era also can be replicated with no real matter field for the AeST theory. The cosmic evolution resulting from both the power-law and de-Sitter solutions also can be obtained.

Accurate and reliable prediction of individual travel mode choices is crucial for developing multi-mode urban transportation systems, conducting transportation planning and formulating traffic demand management strategies. Traditional discrete choice models have dominated the modelling methods for decades yet suffer from strict model assumptions and low prediction accuracy. In recent years, machine learning (ML) models, such as neural networks and boosting models, are widely used by researchers for travel mode choice prediction and have yielded promising results. However, despite the superior prediction performance, a large body of ML methods, especially the branch of neural network models, is also limited by overfitting and tedious model structure determination process. To bridge this gap, this study proposes an enhanced multilayer perceptron (MLP; a neural network) with two hidden layers for travel mode choice prediction; this MLP is enhanced by XGBoost (a boosting method) for feature selection and a grid search method for optimal hidden neurone determination of each hidden layer. The proposed method was trained and tested on a real resident travel diary dataset collected in Chengdu, China., Comment: we need to improve the paper in many ways

As is well-known, black hole plays an important role in probing the quantum effects of gravity in the strong-field regime. In this paper, we focus on a polymerized black hole and investigate the quantum effects on the redshift, blueshift and gravitational redshift of photons emitted by massive particles revolving around this black hole. With a general relativistic formalism, we obtain two elegant and concise expressions for the mass of the black hole and the quantum parameter in terms of few direct observables, and find that all the frequency shifts of photons decrease with the quantum parameter. By expanding the total frequency shift for large orbit of the emitter, low peculiar redshift, and small azimuthal angle, we find that the leading contributions to the frequency shift are originated from the kinematic shift and peculiar shift, and the quantum effects make a negligible contribution. In addition, we study the effects of the plasma on the redshift/blueshift for the first time, and conclude that the presence of plasma results in the decrease of the redshift/blueshift., Comment: 8 pages, 2 figures; published in Physical Review D

Black hole shadow and gravitational lensing play important roles in testing gravitational theories in the strong field regime. As the first-order modifications from quantum gravity, the nonlocality can be manifested by black hole shadow and gravitational lensing. For example, the cut-off parameter introduced by nonlocality will affect the shape and size of the black hole shadow, and also affect the deflection angle of light rays. In this paper, we mainly investigate the effects of the nonlocality on the black hole shadow and the gravitational lensing for two types of rotating black holes in nonlocal gravity. It is found that the size of the black hole shadow decreases with the cut-off parameter since the nonlocality weakens the gravitational constant, and the shape of the shadow gets more deformed with the increase of the cut-off parameter. However, if the rotation parameter is small, the shape of the shadow is almost a circle even though the cut-off parameter approaches its maximum. The energy emission rate in both models is also studied. The results show that there is a peak for each curve and the peak decreases and shifts to the low frequency with the increase of the cut-off parameter. Besides, we also explore the shadow of both types of black holes surrounded by a nonmagnetized pressureless plasma which satisfies the separability condition. It is found that the plasma has a frequency-dependent dispersive effect on the size and shape of the black hole shadow. For the gravitational lensing, we find that the cut-off parameter of model A makes a positive contribution to the deflection angle, which can be compared with the contribution of the rotation parameter, while the cut-off parameter of model B makes a negative contribution which can be ignored. These results may be helpful for probing nonlocal gravity in future observations., Comment: 16 pages, 11 figures

It is well known that general relativity is an effective theory of gravity at low energy scale, and actually quantum effects cannot be ignored in the strong-field regime. As a strong gravitational object, black hole plays a key role in testing the quantum effects of gravity in the strong-field regime. In this paper, we focus on black hole in effective loop quantum gravity and investigate what the influences are of the quantum effects on the weak and strong bending angles of light rays. We find that this black hole could be a Schwarzschild black hole, a regular black hole, a one-way traversable wormhole, or a two-way traversable wormhole for the different values of the quantum parameter, and the strong bending angle for this compact object exhibits two different divergent behaviors, i.e., the logarithmic divergence and non-logarithmic divergence. There are a series of relativistic images on both sides of the optical axis. Only the outermost one can be resolved as a single image, and all others are packed together at the limiting angular position. It is interesting to note that the angular separation between the outermost relativistic image and the others initially increases and then decreases as the quantum parameter increases, indicating that there is a maximum in the angular separation. The maximum is reached after the black hole becomes a wormhole, which can be taken as a signal for the formation of the wormhole. Moreover, the limiting angular position decreases as the quantum parameter increases but a little bounce occurs after the formation of the wormhole, and the relative magnification in magnitudes first decreases and then increases as the quantum parameter increases., Comment: 14 pages, 6 figures; accepted for publication in Physical Review D

Many works have investigated the adversarial attacks or defenses under the settings where a bounded and imperceptible perturbation can be added to the input. However in the real-world, the attacker does not need to comply with this restriction. In fact, more threats to the deep model come from unrestricted adversarial examples, that is, the attacker makes large and visible modifications on the image, which causes the model classifying mistakenly, but does not affect the normal observation in human perspective. Unrestricted adversarial attack is a popular and practical direction but has not been studied thoroughly. We organize this competition with the purpose of exploring more effective unrestricted adversarial attack algorithm, so as to accelerate the academical research on the model robustness under stronger unbounded attacks. The competition is held on the TianChi platform (\url{https://tianchi.aliyun.com/competition/entrance/531853/introduction}) as one of the series of AI Security Challengers Program., Comment: CVPR-2021 AIC Phase VI Track2: Unrestricted Adversarial Attacks on ImageNet

Due to the vulnerability of deep neural networks (DNNs) to adversarial examples, a large number of defense techniques have been proposed to alleviate this problem in recent years. However, the progress of building more robust models is usually hampered by the incomplete or incorrect robustness evaluation. To accelerate the research on reliable evaluation of adversarial robustness of the current defense models in image classification, the TSAIL group at Tsinghua University and the Alibaba Security group organized this competition along with a CVPR 2021 workshop on adversarial machine learning (https://aisecure-workshop.github.io/amlcvpr2021/). The purpose of this competition is to motivate novel attack algorithms to evaluate adversarial robustness more effectively and reliably. The participants were encouraged to develop stronger white-box attack algorithms to find the worst-case robustness of different defenses. This competition was conducted on an adversarial robustness evaluation platform -- ARES (https://github.com/thu-ml/ares), and is held on the TianChi platform (https://tianchi.aliyun.com/competition/entrance/531847/introduction) as one of the series of AI Security Challengers Program. After the competition, we summarized the results and established a new adversarial robustness benchmark at https://ml.cs.tsinghua.edu.cn/ares-bench/, which allows users to upload adversarial attack algorithms and defense models for evaluation., Comment: Competition Report

Deep learning models are vulnerable to adversarial examples, which can fool a target classifier by imposing imperceptible perturbations onto natural examples. In this work, we consider the practical and challenging decision-based black-box adversarial setting, where the attacker can only acquire the final classification labels by querying the target model without access to the model's details. Under this setting, existing works often rely on heuristics and exhibit unsatisfactory performance. To better understand the rationality of these heuristics and the limitations of existing methods, we propose to automatically discover decision-based adversarial attack algorithms. In our approach, we construct a search space using basic mathematical operations as building blocks and develop a random search algorithm to efficiently explore this space by incorporating several pruning techniques and intuitive priors inspired by program synthesis works. Although we use a small and fast model to efficiently evaluate attack algorithms during the search, extensive experiments demonstrate that the discovered algorithms are simple yet query-efficient when transferred to larger normal and defensive models on the CIFAR-10 and ImageNet datasets. They achieve comparable or better performance than the state-of-the-art decision-based attack methods consistently., Comment: 16 pages, 6 figures

In this paper, we study the thick brane system in the so-called $f(Q)$ gravity, where the gravitational interaction was encoded by the nonmetricity $Q$ like scalar curvature $R$ in general relativity. With a special choice of $f(Q)=Q-b Q^n$, we find that the thick brane system can be solved analytically with the first-order formalism, where the complicated second-order differential equation is transformed to several first-order differential equations. Moreover, the stability of the thick brane system under tensor perturbation is also investigated. It is shown that the tachyonic states are absent and the graviton zero mode can be localized on the brane. Thus, the four-dimensional Newtonian potential can be recovered at low energy. Besides, the corrections of the massive graviton Kaluza-Klein modes to the Newtonian potential are also analyzed briefly.

In this paper, we investigate thick branes generated by a scalar field in mimetic gravity theory. By introducing two auxiliary super-potentials, we transform the second-order field equations of the system into a set of first-order equations. With this first-order formalism, several types of analytical thick brane solutions are obtained. Then, tensor and scalar perturbations are analysed. We find that both kinds of perturbations are stable. The effective potentials for the tensor and scalar perturbations are dual to each other. The tensor zero mode can be localized on the brane while the scalar zero mode cannot. Thus, the four-dimensional Newtonian potential can be recovered on the brane., Comment: 9 pages, 6 figures