arXiv subjects

Tong Liu

Publications and source records attributed to Tong Liu.

At least 109 records · Page 6Linked to original sources

arXiv2025

The Art of Hide and Seek: Making Pickle-Based Model Supply Chain Poisoning Stealthy Again

Pickle deserialization vulnerabilities have persisted throughout Python's history, remaining widely recognized yet unresolved. Due to its ability to transparently save and restore complex objects into byte streams, many AI/ML frameworks continue to adopt pickle as the model serialization protocol despite its inherent risks. As the open-source model ecosystem grows, model-sharing platforms such as Hugging Face have attracted massive participation, significantly amplifying the real-world risks of pickle exploitation and opening new avenues for model supply chain poisoning. Although several state-of-the-art scanners have been developed to detect poisoned models, their incomplete understanding of the poisoning surface leaves the detection logic fragile and allows attackers to bypass them. In this work, we present the first systematic disclosure of the pickle-based model poisoning surface from both model loading and risky function perspectives. Our research demonstrates how pickle-based model poisoning can remain stealthy and highlights critical gaps in current scanning solutions. On the model loading surface, we identify 22 distinct pickle-based model loading paths across five foundational AI/ML frameworks, 19 of which are entirely missed by existing scanners. We further develop a bypass technique named Exception-Oriented Programming (EOP) and discover 9 EOP instances, 7 of which can bypass all scanners. On the risky function surface, we discover 133 exploitable gadgets, achieving almost a 100% bypass rate. Even against the best-performing scanner, these gadgets maintain an 89% bypass rate. By systematically revealing the pickle-based model poisoning surface, we achieve practical and robust bypasses against real-world scanners. We responsibly disclose our findings to corresponding vendors, receiving acknowledgments and a $6000 bug bounty.

cs.CR↗

arXiv2025

Variabilities driven by satellite black hole migration in AGN disks

The physical origin of active galactic nucleus (AGN) variability remains unclear. Here we propose that the magnetic reconnection induced by the migration of satellite black holes (sBHs) in the AGN disk can be a new plausible mechanism for AGN short-term variability. During the sBH migration, the co-moving plasmas surrounding the sBH could influence the large-scale magnetic field of the AGN disk and trigger the magnetic reconnections to contribute to AGN UV/optical variability. Meanwhile, high-magnetization plasmas are more likely to escape the disk and cause a secondary magnetic reconnection in the corona. For a $\sim 10^{2}-10^{3}~{M_\mathrm{\odot}}$ sBH in the inner regions of the disk surrounding a supermassive black hole with $\sim 10^{7}~{M_\mathrm{\odot}}$, the reconnection process occurred in the space out of the disk should produce X-ray emission, which can last $\sim 10^3-10^6~\rm s$ with the luminosity $\sim 10^{38}- 10^{42}~\rm{erg ~s^{-1}}$.

astro-ph.HE↗

arXiv2025

Ear-Keeper: A Cross-Platform AI System for Rapid and Accurate Ear Disease Diagnosis

Early and accurate detection systems for ear diseases, powered by deep learning, are essential for preventing hearing impairment and improving population health. However, the limited diversity of existing otoendoscopy datasets and the poor balance between diagnostic accuracy, computational efficiency, and model size have hindered the translation of artificial intelligence (AI) algorithms into healthcare applications. In this study, we constructed a large-scale, multi-center otoendoscopy dataset covering eight common ear diseases and healthy cases. Building upon this resource, we developed Best-EarNet, an ultrafast and lightweight deep learning architecture integrating a novel Local-Global Spatial Feature Fusion Module with a multi-scale supervision strategy, enabling real-time and accurate classification of ear conditions. Leveraging transfer learning, Best-EarNet, with a model size of only 2.94 MB, achieved diagnostic accuracies of 95.23% on an internal test set (22,581 images) and 92.14% on an external test set (1,652 images), while requiring only 0.0125 seconds (80 frames per second) to process a single image on a standard CPU. Further subgroup analysis by gender and age showed consistently excellent performance of Best-EarNet across all demographic groups. To enhance clinical interpretability and user trust, we incorporated Grad-CAM-based visualization, highlighting the specific abnormal ear regions contributing to AI predictions. Most importantly, we developed Ear-Keeper, a cross-platform intelligent diagnosis system built upon Best-EarNet, deployable on smartphones, tablets, and personal computers. Ear-Keeper enables public users and healthcare providers to perform comprehensive real-time video-based ear canal screening, supporting early detection and timely intervention of ear diseases.

cs.CV↗

arXiv2025

FocalPO: Enhancing Preference Optimizing by Focusing on Correct Preference Rankings

Efficient preference optimization algorithms such as Direct Preference Optimization (DPO) have become a popular approach in aligning large language models (LLMs) with human preferences. These algorithms implicitly treat the LLM as a reward model, and focus on training it to correct misranked preference pairs. However, recent work~\citep{chen2024preference} empirically finds that DPO training \textit{rarely improves these misranked preference pairs}, despite its gradient emphasizing on these cases. We introduce FocalPO, a DPO variant that instead \textit{down-weighs} misranked preference pairs and prioritizes enhancing the model's understanding of pairs that it can already rank correctly. Inspired by Focal Loss used in vision tasks, FocalPO achieves this by adding a modulating factor to dynamically scale DPO loss. Our experiment demonstrates that FocalPO surpasses DPO and its variants on popular benchmarks like Alpaca Eval 2.0 using Mistral-Base-7B and Llama-3-Instruct-8B, with the introduced hyperparameter fixed. Additionally, we empirically reveals how FocalPO affects training on correct and incorrect sample groups, further underscoring its effectiveness.

cs.CL↗

arXiv2025

A $6.37\,{\rm Hz}$ quasi-periodic oscillating photospheric emission in GRB~240825A

Using data from Swift-BAT and Fermi-GBM, we report the first detection of a high-confidence quasi-periodic oscillation (QPO) in the thermal emission of gamma-ray burst GRB~240825A. The spectral analysis of the burst reveals two radiation components, including a thermal emission dominant in $100 \text{--} 300\,\mathrm{keV}$ and a non-thermal emission spanning a wide energy range. During the interval $2.07 \text{--} 3.25\,\mathrm{s}$ post-trigger, a strong QPO signal at $6.37\,\mathrm{Hz}$ ($\gtrsim 5σ$ confidence) is identified in the $100 \text{--} 300\,\mathrm{keV}$ thermal-dominated band. The variability analysis of the non-thermal component ($15 \text{--} 30\,\mathrm{keV}$) uncovered a $0.67\,\mathrm{Hz}$ QPO, consistent with light-curve modeling using periodic fast-rise exponential decay pulses. The strong QPO in the photospheric emission directly indicates a quasi-periodic oscillating jet. Together with the non-thermal emission variability, we show that this QPO can be explained in terms of a helical structure in the jet, where the viewing angle to the dominant emission region in the jet undergoes periodic changes.

astro-ph.HE↗

arXiv2025

Anisotropy of core-collapse supernovae effected by AGN disks

Active galactic nucleus (AGN) disk provide dense environments to influence on the star formation, evolution, and migration. In AGN disks, pressure gradients and migration accelerations could create the anisotropy of the core-collapse supernovae (CCSNe) when the massive stars explode at the end of their lives. In this study, we construct the equilibrium equations by considering the above two factors and then compute the light curves for three types of progenitor stars at different locations of AGN disks for the different supermassive black hole (SMBH) masses, accretion efficiencies, explosion energies, and masses of ejecta. The results show that the migration acceleration has more significant effects on the anisotropic explosions than the pressure gradients of the AGN disks. The anisotropic luminosities are pronounced at large radii, and massive SMBHs would suppress the anisotropy and reduce the total luminosity.

astro-ph.HE↗

arXiv2025

User Long-Term Multi-Interest Retrieval Model for Recommendation

User behavior sequence modeling, which captures user interest from rich historical interactions, is pivotal for industrial recommendation systems. Despite breakthroughs in ranking-stage models capable of leveraging ultra-long behavior sequences with length scaling up to thousands, existing retrieval models remain constrained to sequences of hundreds of behaviors due to two main challenges. One is strict latency budget imposed by real-time service over large-scale candidate pool. The other is the absence of target-aware mechanisms and cross-interaction architectures, which prevent utilizing ranking-like techniques to simplify long sequence modeling. To address these limitations, we propose a new framework named User Long-term Multi-Interest Retrieval Model(ULIM), which enables thousand-scale behavior modeling in retrieval stages. ULIM includes two novel components: 1)Category-Aware Hierarchical Dual-Interest Learning partitions long behavior sequences into multiple category-aware subsequences representing multi-interest and jointly optimizes long-term and short-term interests within specific interest cluster. 2)Pointer-Enhanced Cascaded Category-to-Item Retrieval introduces Pointer-Generator Interest Network(PGIN) for next-category prediction, followed by next-item retrieval upon the top-K predicted categories. Comprehensive experiments on Taobao dataset show that ULIM achieves substantial improvement over state-of-the-art methods, and brings 5.54% clicks, 11.01% orders and 4.03% GMV lift for Taobaomiaosha, a notable mini-app of Taobao.

cs.IR↗

arXiv2025

Distributional Soft Actor-Critic with Diffusion Policy

Reinforcement learning has been proven to be highly effective in handling complex control tasks. Traditional methods typically use unimodal distributions, such as Gaussian distributions, to model the output of value distributions. However, unimodal distribution often and easily causes bias in value function estimation, leading to poor algorithm performance. This paper proposes a distributional reinforcement learning algorithm called DSAC-D (Distributed Soft Actor Critic with Diffusion Policy) to address the challenges of estimating bias in value functions and obtaining multimodal policy representations. A multimodal distributional policy iteration framework that can converge to the optimal policy was established by introducing policy entropy and value distribution function. A diffusion value network that can accurately characterize the distribution of multi peaks was constructed by generating a set of reward samples through reverse sampling using a diffusion model. Based on this, a distributional reinforcement learning algorithm with dual diffusion of the value network and the policy network was derived. MuJoCo testing tasks demonstrate that the proposed algorithm not only learns multimodal policy, but also achieves state-of-the-art (SOTA) performance in all 9 control tasks, with significant suppression of estimation bias and total average return improvement of over 10% compared to existing mainstream algorithms. The results of real vehicle testing show that DSAC-D can accurately characterize the multimodal distribution of different driving styles, and the diffusion policy network can characterize multimodal trajectories.

cs.LG↗

arXiv2025

USD: A User-Intent-Driven Sampling and Dual-Debiasing Framework for Large-Scale Homepage Recommendations

Large-scale homepage recommendations face critical challenges from pseudo-negative samples caused by exposure bias, where non-clicks may indicate inattention rather than disinterest. Existing work lacks thorough analysis of invalid exposures and typically addresses isolated aspects (e.g., sampling strategies), overlooking the critical impact of pseudo-positive samples - such as homepage clicks merely to visit marketing portals. We propose a unified framework for large-scale homepage recommendation sampling and debiasing. Our framework consists of two key components: (1) a user intent-aware negative sampling module to filter invalid exposure samples, and (2) an intent-driven dual-debiasing module that jointly corrects exposure bias and click bias. Extensive online experiments on Taobao demonstrate the efficacy of our framework, achieving significant improvements in user click-through rates (UCTR) by 35.4% and 14.5% in two variants of the marketing block on the Taobao homepage, Baiyibutie and Taobaomiaosha.

cs.IR↗

arXiv2025

Layer Decomposition and Morphological Reconstruction for Task-Oriented Infrared Image Enhancement

Infrared image helps improve the perception capabilities of autonomous driving in complex weather conditions such as fog, rain, and low light. However, infrared image often suffers from low contrast, especially in non-heat-emitting targets like bicycles, which significantly affects the performance of downstream high-level vision tasks. Furthermore, achieving contrast enhancement without amplifying noise and losing important information remains a challenge. To address these challenges, we propose a task-oriented infrared image enhancement method. Our approach consists of two key components: layer decomposition and saliency information extraction. First, we design an layer decomposition method for infrared images, which enhances scene details while preserving dark region features, providing more features for subsequent saliency information extraction. Then, we propose a morphological reconstruction-based saliency extraction method that effectively extracts and enhances target information without amplifying noise. Our method improves the image quality for object detection and semantic segmentation tasks. Extensive experiments demonstrate that our approach outperforms state-of-the-art methods.

cs.CV↗

arXiv2025

Cross-modal State Space Modeling for Real-time RGB-thermal Wild Scene Semantic Segmentation

The integration of RGB and thermal data can significantly improve semantic segmentation performance in wild environments for field robots. Nevertheless, multi-source data processing (e.g. Transformer-based approaches) imposes significant computational overhead, presenting challenges for resource-constrained systems. To resolve this critical limitation, we introduced CM-SSM, an efficient RGB-thermal semantic segmentation architecture leveraging a cross-modal state space modeling (SSM) approach. Our framework comprises two key components. First, we introduced a cross-modal 2D-selective-scan (CM-SS2D) module to establish SSM between RGB and thermal modalities, which constructs cross-modal visual sequences and derives hidden state representations of one modality from the other. Second, we developed a cross-modal state space association (CM-SSA) module that effectively integrates global associations from CM-SS2D with local spatial features extracted through convolutional operations. In contrast with Transformer-based approaches, CM-SSM achieves linear computational complexity with respect to image resolution. Experimental results show that CM-SSM achieves state-of-the-art performance on the CART dataset with fewer parameters and lower computational cost. Further experiments on the PST900 dataset demonstrate its generalizability. Codes are available at https://github.com/xiaodonguo/CMSSM.

cs.CV↗

arXiv2025

FrontendBench: A Benchmark for Evaluating LLMs on Front-End Development via Automatic Evaluation

Large Language Models (LLMs) have made significant strides in front-end code generation. However, existing benchmarks exhibit several critical limitations: many tasks are overly simplistic, test cases often lack rigor, and end-to-end validation is absent. These issues hinder the accurate assessment of model performance. To address these challenges, we present FrontendBench, a benchmark co-developed by humans and LLMs. FrontendBench categorizes tasks based on code functionality and incorporates interactive test scenarios, enabling a more comprehensive and practical evaluation of front-end code generation capabilities. The benchmark comprises 148 meticulously crafted prompt-test case pairs spanning five levels of web components, from basic UI elements to complex interactive features. Each task reflects realistic front-end development challenges. Furthermore, we introduce an automatic evaluation framework that executes generated code within a sandbox environment and assesses outcomes using predefined test scripts. This framework achieves a 90.54% agreement rate with expert human evaluations, demonstrating high reliability. We benchmark several state-of-the-art LLMs on FrontendBench and observe substantial performance disparities in handling real-world front-end tasks. These results highlight FrontendBench as a reliable and scalable benchmark, supporting consistent multimodal evaluation and providing a robust foundation for future research in front-end code generation. Our data and code will be released soon.

cs.SE↗

arXiv2025

Upcycling Large Language Models into Mixture of Experts

Upcycling pre-trained dense language models into sparse mixture-of-experts (MoE) models is an efficient approach to increase the model capacity of already trained models. However, optimal techniques for upcycling at scale remain unclear. In this work, we conduct an extensive study of upcycling methods and hyperparameters for billion-parameter scale language models. We propose a novel "virtual group" initialization scheme and weight scaling approach to enable upcycling into fine-grained MoE architectures. Through ablations, we find that upcycling outperforms continued dense model training. In addition, we show that softmax-then-topK expert routing improves over topK-then-softmax approach and higher granularity MoEs can help improve accuracy. Finally, we upcycled Nemotron-4 15B on 1T tokens and compared it to a continuously trained version of the same model on the same 1T tokens: the continuous trained model achieved 65.3% MMLU, whereas the upcycled model achieved 67.6%. Our results offer insights and best practices to effectively leverage upcycling for building MoE language models. Code is available.

cs.CL↗

arXiv2025

Multimodal Pragmatic Jailbreak on Text-to-image Models

Diffusion models have recently achieved remarkable advancements in terms of image quality and fidelity to textual prompts. Concurrently, the safety of such generative models has become an area of growing concern. This work introduces a novel type of jailbreak, which triggers T2I models to generate the image with visual text, where the image and the text, although considered to be safe in isolation, combine to form unsafe content. To systematically explore this phenomenon, we propose a dataset to evaluate the current diffusion-based text-to-image (T2I) models under such jailbreak. We benchmark nine representative T2I models, including two closed-source commercial models. Experimental results reveal a concerning tendency to produce unsafe content: all tested models suffer from such type of jailbreak, with rates of unsafe generation ranging from around 10\% to 70\% where DALLE 3 demonstrates almost the highest unsafety. In real-world scenarios, various filters such as keyword blocklists, customized prompt filters, and NSFW image filters, are commonly employed to mitigate these risks. We evaluate the effectiveness of such filters against our jailbreak and found that, while these filters may be effective for single modality detection, they fail to work against our jailbreak. We also investigate the underlying reason for such jailbreaks, from the perspective of text rendering capability and training data. Our work provides a foundation for further development towards more secure and reliable T2I models. Project page at https://multimodalpragmatic.github.io/.

cs.CV↗

arXiv2025

The FAST Globular Cluster Pulsar Survey (GC FANS)

By January 2025, 60 pulsars were discovered by the Five-hundred-meter Aperture Spherical radio Telescope globular cluster (GC) pulsar survey (GC FANS), with spin periods spanning 1.98 ms to 3960.72 ms. Of these, 55 are millisecond pulsars (MSPs; $P<30$ ms), while 34 are binaries with orbital periods spanning 0.12 days to 466.47 days. This paper describes GC FANS, a deep, thorough search for pulsars in 41 GCs in the FAST sky ($-14^\circ < δ< 65^\circ$) and describes new discoveries in 14 of them. We present updated timing solutions for M92A, NGC 6712A, M71A, and M71E, all of which are ``spider'' pulsars with short orbital periods. We present new timing solutions for M71B, C, and D. With orbital periods of $\sim$466 and 378 days, M71B and M71C are the widest known GC binaries; these systems resemble the normal wide MSP-He WD systems in the Galactic disk. With a spin period of 101 ms, M71D is in an eccentric ($e\sim$0.63) orbit with an 11-day period and a massive companion; the system has a total mass of $2.63 \pm 0.08 \, M_{\odot}$. These features and its large characteristic age suggest it is a double neutron star system (DNS) formed via massive binary evolution early in the cluster's history, akin to Galactic disk DNSs--unlike other candidate GC DNSs, which typically form dynamically. A comparative analysis of GC pulsar populations within FAST's sky reveals that most clusters (10 of 14) resemble the Galactic disk MSP population, likely due to lower stellar densities.

astro-ph.HE↗

arXiv2025

Enhanced quantum sensing in time-modulated non-Hermitian systems

Enhancing the sensitivity of quantum sensing near an exceptional point represents a significant phenomenon in non-Hermitian (NH) systems. However, the application of this property in time-modulated NH systems remains largely unexplored. In this work, we propose two theoretical schemes to achieve enhanced quantum sensing in time-modulated NH systems by leveraging the coalescence of eigenvalues and eigenstates. We conduct a comprehensive analysis of the full energy spectrum, including both real and imaginary components, the population distribution of eigenstates, and various characteristics associated with optimal conditions for sensitivity enhancement. Numerical simulations confirm that eigenvalue-based quantum sensors exhibit a 9.21-fold improvement compared to conventional Hermitian sensors, aligning with the performance of existing time-independent NH sensors. Conversely, for eigenstate-based quantum sensors, the enhancement reaches up to 50 times that of conventional Hermitian sensors, surpassing the results of existing time-independent NH sensors. Moreover, the eigenstate-based sensor exhibits divergent susceptibility even when not close to an exceptional point. Our findings pave the way for advanced sensing in time-sensitive contexts, thereby complementing existing efforts aimed at harnessing the unique properties of open systems.

quant-ph↗

arXiv2025

Efficient conversion from fermionic Gaussian states to matrix product states

Fermionic Gaussian states are eigenstates of quadratic Hamiltonians and are widely used in quantum many-body problems. We propose a highly efficient algorithm that converts fermionic Gaussian states to matrix product states. It can be formulated for finite-size systems without translation invariance, but becomes particularly appealing when applied to infinite systems with translation invariance. If the ground states of a topologically ordered system on infinite cylinders are expressed as matrix product states, then the fixed points of the transfer matrix can be harnessed to filter out the anyon eigenbasis, also known as minimally entangled states. This allows for efficient computation of universal properties, such as the entanglement spectrum and modular matrices. The potential of our method is demonstrated by numerical calculations in two chiral spin liquids that have the same topological orders as the bosonic Laughlin and Moore-Read states, respectively. The anyon eigenbasis for the first one has been worked out before and serves as a useful benchmark. The anyon eigenbasis of the second one is, however, not transparent, and its successful construction provides a nontrivial corroboration of our method.

cond-mat.str-el↗

arXiv2025

Interplay between disorder and topology in Thouless pumping on a superconducting quantum processor

Topological phases are robust against weak perturbations, but break down when disorder becomes sufficiently strong. However, moderate disorder can also induce topologically nontrivial phases. Thouless pumping, as a (1+1)D counterpart of the integer quantum Hall effect, is one of the simplest manifestations of topology. Here, we report experimental observations of the competition and interplay between Thouless pumping and disorder on a 41-qubit superconducting quantum processor. We improve a Floquet engineering technique to realize cycles of adiabatic pumping by simultaneously varying the on-site potentials and the hopping couplings. We demonstrate Thouless pumping in the presence of disorder and show its breakdown as the strength of disorder increases. Moreover, we observe two types of topological pumping that are induced by on-site potential disorder and hopping disorder, respectively. In particular, an intrinsic topological pump that is induced by quasi-periodic hopping disorder has never been experimentally realized before. Our highly controllable system provides a valuable quantum simulating platform for studying various aspects of topological physics in the presence of disorder.

quant-ph↗