arXiv subjects

Haitao Xu

Publications and source records attributed to Haitao Xu.

At least 37 records · Page 2Linked to original sources

arXiv2025

Vul-LMGNNs: Fusing language models and online-distilled graph neural networks for code vulnerability detection

Code Language Models (codeLMs) and Graph Neural Networks (GNNs) are widely used in code vulnerability detection. However, GNNs often rely on aggregating information from adjacent nodes, limiting structural information propagation across layers. While codeLMs can supplement GNNs with semantic information, existing integration methods underexplore their collaborative potential. To address these challenges, we propose Vul-LMGNNs, integrating pre-trained codeLMs with GNNs to enable cross-layer propagation of semantic and structural information. Vul-LMGNNs leverage Code Property Graphs (CPGs) to incorporate syntax, control flow, and data dependencies, using gated GNNs for structural extraction. An online knowledge distillation (KD) mechanism allows a student GNN to capture structural information from a trained counterpart via alternating training. Additionally, an "implicit-explicit" joint training framework leverages codeLMs to initialize embeddings and propagate code semantics. In the explicit phase, it performs late fusion via linear interpolation. Evaluations on real-world vulnerability datasets show Vul-LMGNNs outperform 17 state-of-the-art approaches. Source code is available at: https://github.com/Vul-LMGNN/vul-LMGNN.

cs.CR↗

arXiv2025

From Image- to Pixel-level: Label-efficient Hyperspectral Image Reconstruction

Current hyperspectral image (HSI) reconstruction methods primarily rely on image-level approaches, which are time-consuming to form abundant high-quality HSIs through imagers. In contrast, spectrometers offer a more efficient alternative by capturing high-fidelity point spectra, enabling pixel-level HSI reconstruction that balances accuracy and label efficiency. To this end, we introduce a pixel-level spectral super-resolution (Pixel-SSR) paradigm that reconstructs HSI from RGB and point spectra. Despite its advantages, Pixel-SSR presents two key challenges: 1) generalizability to novel scenes lacking point spectra, and 2) effective information extraction to promote reconstruction accuracy. To address the first challenge, a Gamma-modeled strategy is investigated to synthesize point spectra based on their intrinsic properties, including nonnegativity, a skewed distribution, and a positive correlation. Furthermore, complementary three-branch prompts from RGB and point spectra are extracted with a Dynamic Prompt Mamba (DyPro-Mamba), which progressively directs the reconstruction with global spatial distributions, edge details, and spectral dependency. Comprehensive evaluations, including horizontal comparisons with leading methods and vertical assessments across unsupervised and image-level supervised paradigms, demonstrate that ours achieves competitive reconstruction accuracy with efficient label consumption.

cs.CV↗

arXiv2025

The contribution of dilatational motion to energy flux in homogeneous compressible turbulence

We analyze the energy flux in compressible turbulence by generalizing the exact decomposition recently proposed by Johnson (Phys. Rev. Lett., vol. 124, 2020. 104501) to study incompressible turbulent flows. This allows us to characterize the effect of dilatational motion on the inter-scale energy transfer in three-dimensional compressible turbulence. Our analysis reveals that the contribution of dilatational motion to energy transfer is due to three different physical mechanisms: the interaction between dilatation and strain, between dilatation and vorticity, and the self-interaction of dilatational motion across scales. By analyzing numerical simulations of flows at moderate turbulent Mach numbers ($Ma_t \lesssim 0.3$), we validate our theoretical derivations and provide a quantitative description of the role of dilatational motion in energy transfer. In particular, we determine the scaling dependence of the dilatational contributions on the turbulent Mach number. Moreover, our findings reveal that the eddy-viscosity assumption often used in large-eddy simulations, in the spirit of the approach used for incompressible flows, effectively neglects the interaction between solenoidal-dilatational energy transfer and overestimate dilatational effects.

physics.flu-dyn↗

arXiv2025

The Two-Center Problem of Uncertain Points on Cactus Graphs

We study the two-center problem on cactus graphs in facility locations, which aims to place two facilities on the graph network to serve customers in order to minimize the maximum transportation cost. In our problem, the location of each customer is uncertain and may appear at $O(m)$ points on the network with probabilities. More specifically, given are a cactus graph $G$ and a set $\calP$ of $n$ (weighted) uncertain points where every uncertain point has $O(m)$ possible locations on $G$ each associated with a probability and is of a non-negative weight. The problem aims to compute two centers (points) on $G$ so that the maximum (weighted) expected distance of the $n$ uncertain points to their own expected closest center is minimized. No previous algorithms are known for this problem. In this paper, we present the first algorithm for this problem and it solves the problem in $O(|G|+ m^{2}n^{2}\log mn)$ time.

cs.DS↗

arXiv2024

The Two-Center Problem of Uncertain Points on Trees

In this paper, we consider the (weighted) two-center problem of uncertain points on a tree. Given are a tree $T$ and a set $\calP$ of $n$ (weighted) uncertain points each of which has $m$ possible locations on $T$ associated with probabilities. The goal is to compute two points on $T$, i.e., two centers with respect to $\calP$, so that the maximum (weighted) expected distance of $n$ uncertain points to their own expected closest center is minimized. This problem can be solved in $O(|T|+ n^{2}\log n\log mn + mn\log^2 mn \log n)$ time by the algorithm for the general $k$-center problem. In this paper, we give a more efficient and simple algorithm that solves this problem in $O(|T| + mn\log mn)$ time.

cs.DS↗

arXiv2024

Unimolecular Chemical Kinetics in the Interstellar: Competition of Infrared Radiation and Collision Activation Mechanisms

Unimolecular gas phase chemical reactions could be activated by both infrared (IR) radiation and inter-molecular collision in the interstellar environment. Understanding the interplay and competition between the radiation and collision activation mechanisms is crucial for assessing accurate reaction rate constants with an appropriate model. In this work, guided by an extended version of Lindemann theory that considers the contribution of both the radiation-activation and collision-activation to the rate constant of unimolecular reactions, we show that the relative importance of the two mechanisms can be measured by a dimensionless number $PR$ that is the ratio of the collision frequency to the radiation absorption rate of the molecule. The reaction kinetic is dominated by collision-activation or by radiation activation depending on whether $PR$ is larger or smaller than a reference value $PR^*$, which is determined to be $PR^* \approx 10$ based on magnitudes of molecular properties and is verified by detailed calculations of a number of typical interstellar unimolecular reactions. This method of evaluating the relative importance of the two mechanisms is checked against master equation calculations of two interstellar reactions: the dissociation reaction of silicilic acid around the asymptotic giant branch (AGB) star and the methyl association in Titan's atmosphere, and the validity is verified. The method can be used in the future to help determine the appropriate and effective modeling approach for chemical reactions in astrophysical environments.

astro-ph.GA↗

arXiv2024

The USTC-NERCSLIP Systems for the CHiME-8 NOTSOFAR-1 Challenge

This technical report outlines our submission system for the CHiME-8 NOTSOFAR-1 Challenge. The primary difficulty of this challenge is the dataset recorded across various conference rooms, which captures real-world complexities such as high overlap rates, background noises, a variable number of speakers, and natural conversation styles. To address these issues, we optimized the system in several aspects: For front-end speech signal processing, we introduced a data-driven joint training method for diarization and separation (JDS) to enhance audio quality. Additionally, we also integrated traditional guided source separation (GSS) for multi-channel track to provide complementary information for the JDS. For back-end speech recognition, we enhanced Whisper with WavLM, ConvNeXt, and Transformer innovations, applying multi-task training and Noise KLD augmentation, to significantly advance ASR robustness and accuracy. Our system attained a Time-Constrained minimum Permutation Word Error Rate (tcpWER) of 14.265% and 22.989% on the CHiME-8 NOTSOFAR-1 Dev-set-2 multi-channel and single-channel tracks, respectively.

eess.AS↗

arXiv2024

High-Temperature Non-Equilibrium Atom-Diatom Collisional Energy Transfer

The change of the vibrational energy within a molecule after collisions with another molecule plays an essential role in the evolution of molecular internal energy distributions, which is also the limiting process in the relaxation of the gas towards equilibrium. Here we investigate the energy transfer between the translational motion and the vibrational motion of the diatom during the atom-diatom collision, the simplest case involving the transfer between inter-molecular and intra-molecular energies. We are interested in the situation when the translational temperature of the gas is high, in which case there are significant probabilities for the vibrational energy to change over widely separated energy levels after a collision. Data from quasi-classical trajectory simulations of the N+N$_2$ system with \textit{ab initio} potential energies suggest that the transition probability dependence on the collisional energy possesses an ``activation-saturation'' behavior and can be described by a simple model. The model allows for explicit evaluation of the vibrational state-to-state transition rate coefficients, from which the evolution of the vibrational energy distribution from any initial conditions can be solved by the master equation approach. An example of the vibrational energy relaxation in the N+N$_2$ system mimicking the gas behind strong shocks in a hypersonic flow is shown and the results are in good agreement with available data.

physics.chem-ph↗

arXiv2024

Marlin: Knowledge-Driven Analysis of Provenance Graphs for Efficient and Robust Detection of Cyber Attacks

Recent research in both academia and industry has validated the effectiveness of provenance graph-based detection for advanced cyber attack detection and investigation. However, analyzing large-scale provenance graphs often results in substantial overhead. To improve performance, existing detection systems implement various optimization strategies. Yet, as several recent studies suggest, these strategies could lose necessary context information and be vulnerable to evasions. Designing a detection system that is efficient and robust against adversarial attacks is an open problem. We introduce Marlin, which approaches cyber attack detection through real-time provenance graph alignment.By leveraging query graphs embedded with attack knowledge, Marlin can efficiently identify entities and events within provenance graphs, embedding targeted analysis and significantly narrowing the search space. Moreover, we incorporate our graph alignment algorithm into a tag propagation-based schema to eliminate the need for storing and reprocessing raw logs. This design significantly reduces in-memory storage requirements and minimizes data processing overhead. As a result, it enables real-time graph alignment while preserving essential context information, thereby enhancing the robustness of cyber attack detection. Moreover, Marlin allows analysts to customize attack query graphs flexibly to detect extended attacks and provide interpretable detection results. We conduct experimental evaluations on two large-scale public datasets containing 257.42 GB of logs and 12 query graphs of varying sizes, covering multiple attack techniques and scenarios. The results show that Marlin can process 137K events per second while accurately identifying 120 subgraphs with 31 confirmed attacks, along with only 1 false positive, demonstrating its efficiency and accuracy in handling massive data.

cs.CR↗

arXiv2024

Struggle with Adversarial Defense? Try Diffusion

Adversarial attacks induce misclassification by introducing subtle perturbations. Recently, diffusion models are applied to the image classifiers to improve adversarial robustness through adversarial training or by purifying adversarial noise. However, diffusion-based adversarial training often encounters convergence challenges and high computational expenses. Additionally, diffusion-based purification inevitably causes data shift and is deemed susceptible to stronger adaptive attacks. To tackle these issues, we propose the Truth Maximization Diffusion Classifier (TMDC), a generative Bayesian classifier that builds upon pre-trained diffusion models and the Bayesian theorem. Unlike data-driven classifiers, TMDC, guided by Bayesian principles, utilizes the conditional likelihood from diffusion models to determine the class probabilities of input images, thereby insulating against the influences of data shift and the limitations of adversarial training. Moreover, to enhance TMDC's resilience against more potent adversarial attacks, we propose an optimization strategy for diffusion classifiers. This strategy involves post-training the diffusion model on perturbed datasets with ground-truth labels as conditions, guiding the diffusion model to learn the data distribution and maximizing the likelihood under the ground-truth labels. The proposed method achieves state-of-the-art performance on the CIFAR10 dataset against heavy white-box attacks and strong adaptive attacks. Specifically, TMDC achieves robust accuracies of 82.81% against $l_{\infty}$ norm-bounded perturbations and 86.05% against $l_{2}$ norm-bounded perturbations, respectively, with $ε=0.05$.

cs.CV↗

arXiv2023

Fed-urlBERT: Client-side Lightweight Federated Transformers for URL Threat Analysis

In evolving cyber landscapes, the detection of malicious URLs calls for cooperation and knowledge sharing across domains. However, collaboration is often hindered by concerns over privacy and business sensitivities. Federated learning addresses these issues by enabling multi-clients collaboration without direct data exchange. Unfortunately, if highly expressive Transformer models are used, clients may face intolerable computational burdens, and the exchange of weights could quickly deplete network bandwidth. In this paper, we propose Fed-urlBERT, a federated URL pre-trained model designed to address both privacy concerns and the need for cross-domain collaboration in cybersecurity. Fed-urlBERT leverages split learning to divide the pre-training model into client and server part, so that the client part takes up less extensive computation resources and bandwidth. Our appraoch achieves performance comparable to centralized model under both independently and identically distributed (IID) and two non-IID data scenarios. Significantly, our federated model shows about an 7% decrease in the FPR compared to the centralized model. Additionally, we implement an adaptive local aggregation strategy that mitigates heterogeneity among clients, demonstrating promising performance improvements. Overall, our study validates the applicability of the proposed Transformer federated learning for URL threat analysis, establishing a foundation for real-world collaborative cybersecurity efforts. The source code is accessible at https://github.com/Davidup1/FedURLBERT.

cs.CR↗

arXiv2023

SparseByteNN: A Novel Mobile Inference Acceleration Framework Based on Fine-Grained Group Sparsity

To address the challenge of increasing network size, researchers have developed sparse models through network pruning. However, maintaining model accuracy while achieving significant speedups on general computing devices remains an open problem. In this paper, we present a novel mobile inference acceleration framework SparseByteNN, which leverages fine-grained kernel sparsity to achieve real-time execution as well as high accuracy. Our framework consists of two parts: (a) A fine-grained kernel sparsity schema with a sparsity granularity between structured pruning and unstructured pruning. It designs multiple sparse patterns for different operators. Combined with our proposed whole network rearrangement strategy, the schema achieves a high compression rate and high precision at the same time. (b) Inference engine co-optimized with the sparse pattern. The conventional wisdom is that this reduction in theoretical FLOPs does not translate into real-world efficiency gains. We aim to correct this misconception by introducing a family of efficient sparse kernels for ARM and WebAssembly. Equipped with our efficient implementation of sparse primitives, we show that sparse versions of MobileNet-v1 outperform strong dense baselines on the efficiency-accuracy curve. Experimental results on Qualcomm 855 show that for 30% sparse MobileNet-v1, SparseByteNN achieves 1.27x speedup over the dense version and 1.29x speedup over the state-of-the-art sparse inference engine MNN with a slight accuracy drop of 0.224%. The source code of SparseByteNN will be available at https://github.com/lswzjuer/SparseByteNN

cs.AI↗

arXiv2023

Emergence of purely nonlinear localized states with frequencies exited from spectral bands

In this work, we revisit the classic model of diatomic chain with cubic nonlinearity and investigate the formation mechanism of nonlinear localized time-periodic solutions (breathers) with frequencies exited the spectral bands. First we employ the long-chain limit to obtain estimates of linear eigenstates, especially those with frequencies near the band edges. As the strength of nonlinearity grows, some frequencies can cross the band edges to turn isolated while the corresponding states gradually change from non-localized to localized. Based on the estimates of linear eigenstates, we derive analytical approximations of these nonlinear states and prove their validity for frequencies within the bands. Moreover, the process of states growing localized are illustrated in both analytical and numerical approaches. Although here we place emphasis on nonlinear middle-localized states with the most generic boundary conditions, the results can also be extended to a wider range of localized states including edge states.

nlin.PS↗

arXiv2022

Low-order moments of the velocity gradient in homogeneous compressible turbulence

We derive from first principles analytic relations for the second and third order moments of the velocity gradient mij = dui/dxj in compressible turbulence, which generalize known relations in incompressible flows. These relations, although derived for homogeneous flows, hold approximately for a mixing layer. We also discuss how to apply these relations to determine all the second and third moments of the velocity gradient experimentally.

physics.flu-dyn↗

arXiv2021

Experimental observation of the elastic range scaling in turbulent flow with polymer additives

Minute amount of long chain flexible polymer dissolved in a turbulent flow can drastically change flow properties, such as reducing the drag and enhancing mixing. One fundamental riddle is how these polymer additives interact with the eddies of different spatial scales existing in the turbulent flow and in turn alter the turbulence energy transfer. Here we show how turbulent kinetic energy is transferred through deferent scales in the presence of the polymer additives. In particular, we observed experimentally the emerging of a new scaling range, referred to as the elastic range, where increasing amount of energy is transferred by the elasticity of the polymers. In addition, the existence of the elastic range prescribes the scaling of high-order velocity statistics. Our findings have important implications to many turbulence systems such as turbulence in plasmas or superfluids where interaction between turbulent eddies and other nonlinear physical mechanisms are often involved.

physics.flu-dyn↗

arXiv2020

Stability of topological edge states under strong nonlinear effects

We examine the role of strong nonlinearity on the topologically-robust edge state in a one-dimensional system. We consider a chain inspired from the Su-Schrieffer-Heeger model, but with a finite-frequency edge state and the dynamics governed by second-order differential equations. We introduce a cubic onsite-nonlinearity and study this nonlinear effect on the edge state's frequency and linear stability. Nonlinear continuation reveals that the edge state loses its typical shape enforced by the chiral symmetry and becomes generally unstable due to various types of instabilities that we analyze using a combination of spectral stability and Krein signature analysis. This results in an initially-excited nonlinear-edge state shedding its energy into the bulk over a long time. However, the stability trends differ both qualitatively and quantitatively when softening and stiffening types of nonlinearity are considered. In the latter, we find a frequency regime where nonlinear edge states can be linearly stable. This enables high-amplitude edge states to remain spatially localized without shedding their energy, a feature that we have confirmed via long-time dynamical simulations. Finally, we examine the robustness of frequency and stability of nonlinear edge states against disorder, and find that those are more robust under a chiral disorder compared to a non-chiral disorder. Moreover, the frequency-regime where high-amplitude edge states were found to be linearly stable remains intact in the presence of small amount of disorder of both types.

nlin.PS↗

arXiv2020

MIME: Mutual Information Minimisation Exploration

We show that reinforcement learning agents that learn by surprise (surprisal) get stuck at abrupt environmental transition boundaries because these transitions are difficult to learn. We propose a counter-intuitive solution that we call Mutual Information Minimising Exploration (MIME) where an agent learns a latent representation of the environment without trying to predict the future states. We show that our agent performs significantly better over sharp transition boundaries while matching the performance of surprisal driven agents elsewhere. In particular, we show state-of-the-art performance on difficult learning games such as Gravitar, Montezuma's Revenge and Doom.

cs.LG↗

arXiv2019

VASE: Variational Assorted Surprise Exploration for Reinforcement Learning

Exploration in environments with continuous control and sparse rewards remains a key challenge in reinforcement learning (RL). Recently, surprise has been used as an intrinsic reward that encourages systematic and efficient exploration. We introduce a new definition of surprise and its RL implementation named Variational Assorted Surprise Exploration (VASE). VASE uses a Bayesian neural network as a model of the environment dynamics and is trained using variational inference, alternately updating the accuracy of the agent's model and policy. Our experiments show that in continuous control sparse reward environments VASE outperforms other surprise-based exploration techniques.

cs.LG↗