arXiv subjects

Songze Li

Publications and source records attributed to Songze Li.

At least 91 records · Page 5Linked to original sources

arXiv2025

TimeSuite: Improving MLLMs for Long Video Understanding via Grounded Tuning

Multimodal Large Language Models (MLLMs) have demonstrated impressive performance in short video understanding. However, understanding long-form videos still remains challenging for MLLMs. This paper proposes TimeSuite, a collection of new designs to adapt the existing short-form video MLLMs for long video understanding, including a simple yet efficient framework to process long video sequence, a high-quality video dataset for grounded tuning of MLLMs, and a carefully-designed instruction tuning task to explicitly incorporate the grounding supervision in the traditional QA format. Specifically, based on VideoChat, we propose our long-video MLLM, coined as VideoChat-T, by implementing a token shuffling to compress long video tokens and introducing Temporal Adaptive Position Encoding (TAPE) to enhance the temporal awareness of visual representation. Meanwhile, we introduce the TimePro, a comprehensive grounding-centric instruction tuning dataset composed of 9 tasks and 349k high-quality grounded annotations. Notably, we design a new instruction tuning task type, called Temporal Grounded Caption, to peform detailed video descriptions with the corresponding time stamps prediction. This explicit temporal location prediction will guide MLLM to correctly attend on the visual content when generating description, and thus reduce the hallucination risk caused by the LLMs. Experimental results demonstrate that our TimeSuite provides a successful solution to enhance the long video understanding capability of short-form MLLM, achieving improvement of 5.6% and 6.8% on the benchmarks of Egoschema and VideoMME, respectively. In addition, VideoChat-T exhibits robust zero-shot temporal grounding capabilities, significantly outperforming the existing state-of-the-art MLLMs. After fine-tuning, it performs on par with the traditional supervised expert models.

cs.CV↗

arXiv2025

When Data Manipulation Meets Attack Goals: An In-depth Survey of Attacks for VLMs

Vision-Language Models (VLMs) have gained considerable prominence in recent years due to their remarkable capability to effectively integrate and process both textual and visual information. This integration has significantly enhanced performance across a diverse spectrum of applications, such as scene perception and robotics. However, the deployment of VLMs has also given rise to critical safety and security concerns, necessitating extensive research to assess the potential vulnerabilities these VLM systems may harbor. In this work, we present an in-depth survey of the attack strategies tailored for VLMs. We categorize these attacks based on their underlying objectives - namely jailbreak, camouflage, and exploitation - while also detailing the various methodologies employed for data manipulation of VLMs. Meanwhile, we outline corresponding defense mechanisms that have been proposed to mitigate these vulnerabilities. By discerning key connections and distinctions among the diverse types of attacks, we propose a compelling taxonomy for VLM attacks. Moreover, we summarize the evaluation metrics that comprehensively describe the characteristics and impact of different attacks on VLMs. Finally, we conclude with a discussion of promising future research directions that could further enhance the robustness and safety of VLMs, emphasizing the importance of ongoing exploration in this critical area of study. To facilitate community engagement, we maintain an up-to-date project page, accessible at: https://github.com/AobtDai/VLM_Attack_Paper_List.

cs.CV↗

arXiv2025

Arbitrary-Threshold Fully Homomorphic Encryption with Lower Complexity

Threshold fully homomorphic encryption (ThFHE) enables multiple parties to compute functions over their sensitive data without leaking data privacy. Most of existing ThFHE schemes are restricted to full threshold and require the participation of \textit{all} parties to output computing results. Compared with these full-threshold schemes, arbitrary threshold (ATh)-FHE schemes are robust to non-participants and can be a promising solution to many real-world applications. However, existing AThFHE schemes are either inefficient to be applied with a large number of parties $N$ and a large data size $K$, or insufficient to tolerate all types of non-participants. In this paper, we propose an AThFHE scheme to handle all types of non-participants with lower complexity over existing schemes. At the core of our scheme is the reduction from AThFHE construction to the design of a new primitive called \textit{approximate secret sharing} (ApproxSS). Particularly, we formulate ApproxSS and prove the correctness and security of AThFHE on top of arbitrary-threshold (ATh)-ApproxSS's properties. Such a reduction reveals that existing AThFHE schemes implicitly design ATh-ApproxSS following a similar idea called ``noisy share''. Nonetheless, their ATh-ApproxSS design has high complexity and become the performance bottleneck. By developing ATASSES, an ATh-ApproxSS scheme based on a novel ``encrypted share'' idea, we reduce the computation (resp. communication) complexity from $\mathcal{O}(N^2K)$ to $\mathcal{O}(N^2+K)$ (resp. from $\mathcal{O}(NK)$ to $\mathcal{O}(N+K)$). We not only theoretically prove the (approximate) correctness and security of ATASSES, but also empirically evaluate its efficiency against existing baselines. Particularly, when applying to a system with one thousand parties, ATASSES achieves a speedup of $3.83\times$ -- $15.4\times$ over baselines.

cs.CR↗

arXiv2025

Rethinking Knowledge in Distillation: An In-context Sample Retrieval Perspective

Conventional knowledge distillation (KD) approaches are designed for the student model to predict similar output as the teacher model for each sample. Unfortunately, the relationship across samples with same class is often neglected. In this paper, we explore to redefine the knowledge in distillation, capturing the relationship between each sample and its corresponding in-context samples (a group of similar samples with the same or different classes), and perform KD from an in-context sample retrieval perspective. As KD is a type of learned label smoothing regularization (LSR), we first conduct a theoretical analysis showing that the teacher's knowledge from the in-context samples is a crucial contributor to regularize the student training with the corresponding samples. Buttressed by the analysis, we propose a novel in-context knowledge distillation (IC-KD) framework that shows its superiority across diverse KD paradigms (offline, online, and teacher-free KD). Firstly, we construct a feature memory bank from the teacher model and retrieve in-context samples for each corresponding sample through retrieval-based learning. We then introduce Positive In-Context Distillation (PICD) to reduce the discrepancy between a sample from the student and the aggregated in-context samples with the same class from the teacher in the logit space. Moreover, Negative In-Context Distillation (NICD) is introduced to separate a sample from the student and the in-context samples with different classes from the teacher in the logit space. Extensive experiments demonstrate that IC-KD is effective across various types of KD, and consistently achieves state-of-the-art performance on CIFAR-100 and ImageNet datasets.

cs.CV↗

arXiv2024

URVFL: Undetectable Data Reconstruction Attack on Vertical Federated Learning

Launching effective malicious attacks in VFL presents unique challenges: 1) Firstly, given the distributed nature of clients' data features and models, each client rigorously guards its privacy and prohibits direct querying, complicating any attempts to steal data; 2) Existing malicious attacks alter the underlying VFL training task, and are hence easily detected by comparing the received gradients with the ones received in honest training. To overcome these challenges, we develop URVFL, a novel attack strategy that evades current detection mechanisms. The key idea is to integrate a discriminator with auxiliary classifier that takes a full advantage of the label information and generates malicious gradients to the victim clients: on one hand, label information helps to better characterize embeddings of samples from distinct classes, yielding an improved reconstruction performance; on the other hand, computing malicious gradients with label information better mimics the honest training, making the malicious gradients indistinguishable from the honest ones, and the attack much more stealthy. Our comprehensive experiments demonstrate that URVFL significantly outperforms existing attacks, and successfully circumvents SOTA detection methods for malicious attacks. Additional ablation studies and evaluations on defenses further underscore the robustness and effectiveness of URVFL. Our code will be available at https://github.com/duanyiyao/URVFL.

cs.LG↗

arXiv2024

Hidden Data Privacy Breaches in Federated Learning

Federated Learning (FL) emerged as a paradigm for conducting machine learning across broad and decentralized datasets, promising enhanced privacy by obviating the need for direct data sharing. However, recent studies show that attackers can steal private data through model manipulation or gradient analysis. Existing attacks are constrained by low theft quantity or low-resolution data, and they are often detected through anomaly monitoring in gradients or weights. In this paper, we propose a novel data-reconstruction attack leveraging malicious code injection, supported by two key techniques, i.e., distinctive and sparse encoding design and block partitioning. Unlike conventional methods that require detectable changes to the model, our method stealthily embeds a hidden model using parameter sharing to systematically extract sensitive data. The Fibonacci-based index design ensures efficient, structured retrieval of memorized data, while the block partitioning method enhances our method's capability to handle high-resolution images by dividing them into smaller, manageable units. Extensive experiments on 4 datasets confirmed that our method is superior to the five state-of-the-art data-reconstruction attacks under the five respective detection methods. Our method can handle large-scale and high-resolution data without being detected or mitigated by state-of-the-art data reconstruction defense methods. In contrast to baselines, our method can be directly applied to both FedAVG and FedSGD scenarios, underscoring the need for developers to devise new defenses against such vulnerabilities. We will open-source our code upon acceptance.

cs.CL↗

arXiv2024

Navigating the Risks: A Survey of Security, Privacy, and Ethics Threats in LLM-Based Agents

With the continuous development of large language models (LLMs), transformer-based models have made groundbreaking advances in numerous natural language processing (NLP) tasks, leading to the emergence of a series of agents that use LLMs as their control hub. While LLMs have achieved success in various tasks, they face numerous security and privacy threats, which become even more severe in the agent scenarios. To enhance the reliability of LLM-based applications, a range of research has emerged to assess and mitigate these risks from different perspectives. To help researchers gain a comprehensive understanding of various risks, this survey collects and analyzes the different threats faced by these agents. To address the challenges posed by previous taxonomies in handling cross-module and cross-stage threats, we propose a novel taxonomy framework based on the sources and impacts. Additionally, we identify six key features of LLM-based agents, based on which we summarize the current research progress and analyze their limitations. Subsequently, we select four representative agents as case studies to analyze the risks they may face in practical use. Finally, based on the aforementioned analyses, we propose future research directions from the perspectives of data, methodology, and policy, respectively.

cs.AI↗

arXiv2024

Spintwistronics: Photonic bilayer topological lattices tuning extreme spin-orbit interactions

Twistronics, the manipulation of Moiré superlattices via the twisting of two layers of two-dimensional (2D) materials to control diverse and nontrivial properties, has recently revolutionized the condensed matter and materials physics. Here, we introduce the principles of twistronics to spin photonics, coining this emerging field spintwistronics. In spintwistronics, instead of 2D materials, the two layers consist of photonic topological spin lattices on a surface plasmonic polariton (SPP) platform. Each 2D SPP wave supports the construction of topological lattices formed by photonic spins with stable skyrmion topology governed by rotational symmetry. By introducing spintwistronics into plasmonics, we demonstrate theoretically and experimentally that two layers of photonic spin lattices can produce Moiré spin superlattices at specific magic angles. These superlattices, modulated periodically by the quantum number of total angular momentum, exhibit novel properties-including new quasiparticle topologies, multiple fractal patterns, extremely slow-light control, and more-that cannot be achieved in conventional plasmonic systems. As a result, they open up multiple degrees of freedom for practical applications in quantum information, optical data storage and chiral light-matter interactions.

physics.optics↗

arXiv2024

Generalized Lagrange Coded Computing: A Flexible Computation-Communication Tradeoff for Resilient, Secure, and Private Computation

We consider the problem of evaluating arbitrary multivariate polynomials over a massive dataset containing multiple inputs, on a distributed computing system with a master node and multiple worker nodes. Generalized Lagrange Coded Computing (GLCC) codes are proposed to simultaneously provide resiliency against stragglers who do not return computation results in time, security against adversarial workers who deliberately modify results for their benefit, and information-theoretic privacy of the dataset amidst possible collusion of workers. GLCC codes are constructed by first partitioning the dataset into multiple groups, then encoding the dataset using carefully designed interpolating polynomials, and sharing multiple encoded data points to each worker, such that interference computation results across groups can be eliminated at the master. Particularly, GLCC codes include the state-of-the-art Lagrange Coded Computing (LCC) codes as a special case, and exhibit a more flexible tradeoff between communication and computation overheads in optimizing system efficiency. Furthermore, we apply GLCC to distributed training of machine learning models, and demonstrate that GLCC codes achieve a speedup of up to $2.5\text{--}3.9\times$ over LCC codes in training time, across experiments for training image classifiers on different datasets, model architectures, and straggler patterns.

cs.IT↗

arXiv2024

InternVideo2: Scaling Foundation Models for Multimodal Video Understanding

We introduce InternVideo2, a new family of video foundation models (ViFM) that achieve the state-of-the-art results in video recognition, video-text tasks, and video-centric dialogue. Our core design is a progressive training approach that unifies the masked video modeling, crossmodal contrastive learning, and next token prediction, scaling up the video encoder size to 6B parameters. At the data level, we prioritize spatiotemporal consistency by semantically segmenting videos and generating video-audio-speech captions. This improves the alignment between video and text. Through extensive experiments, we validate our designs and demonstrate superior performance on over 60 video and audio tasks. Notably, our model outperforms others on various video-related dialogue and long video understanding benchmarks, highlighting its ability to reason and comprehend longer contexts. Code and models are available at https://github.com/OpenGVLab/InternVideo/tree/main/InternVideo2/.

cs.CV↗

arXiv2024

Constructing Adversarial Examples for Vertical Federated Learning: Optimal Client Corruption through Multi-Armed Bandit

Vertical federated learning (VFL), where each participating client holds a subset of data features, has found numerous applications in finance, healthcare, and IoT systems. However, adversarial attacks, particularly through the injection of adversarial examples (AEs), pose serious challenges to the security of VFL models. In this paper, we investigate such vulnerabilities through developing a novel attack to disrupt the VFL inference process, under a practical scenario where the adversary is able to adaptively corrupt a subset of clients. We formulate the problem of finding optimal attack strategies as an online optimization problem, which is decomposed into an inner problem of adversarial example generation (AEG) and an outer problem of corruption pattern selection (CPS). Specifically, we establish the equivalence between the formulated CPS problem and a multi-armed bandit (MAB) problem, and propose the Thompson sampling with Empirical maximum reward (E-TS) algorithm for the adversary to efficiently identify the optimal subset of clients for corruption. The key idea of E-TS is to introduce an estimation of the expected maximum reward for each arm, which helps to specify a small set of competitive arms, on which the exploration for the optimal arm is performed. This significantly reduces the exploration space, which otherwise can quickly become prohibitively large as the number of clients increases. We analytically characterize the regret bound of E-TS, and empirically demonstrate its capability of efficiently revealing the optimal corruption pattern with the highest attack success rate, under various datasets of popular VFL tasks.

cs.LG↗

arXiv2024

Manifoldchain: Maximizing Blockchain Throughput via Bandwidth-Clustered Sharding

Bandwidth limitation is the major bottleneck that hinders scaling throughput of proof-of-work blockchains. To guarantee security, the mining rate of the blockchain is determined by the miners with the lowest bandwidth, resulting in an inefficient bandwidth utilization among fast miners. We propose Manifoldchain, an innovative blockchain sharding protocol that alleviates the impact of slow miners to maximize blockchain throughput. Manifoldchain utilizes a bandwidth-clustered shard formation mechanism that groups miners with similar bandwidths into the same shard. Consequently, this approach enables us to set an optimal mining rate for each shard based on its bandwidth, effectively reducing the waiting time caused by slow miners. Nevertheless, the adversary could corrupt miners with similar bandwidths, thereby concentrating hashing power and potentially creating an adversarial majority within a single shard. To counter this adversarial strategy, we introduce sharing mining, allowing the honest mining power of the entire network to participate in the secure ledger formation of each shard, thereby achieving the same level of security as an unsharded blockchain. Additionally, we introduce an asynchronous atomic commitment mechanism to ensure transaction atomicity across shards with various mining rates. Our theoretical analysis demonstrates that Manifoldchain scales linearly in throughput with the increase in shard numbers and inversely with network delay in each shard. We implement a full system prototype of Manifoldchain, comprehensively evaluated on both simulated and real-world testbeds. These experiments validate its vertical scalability with network bandwidth and horizontal scalability with network size, achieving a substantial improvement of 186% in throughput over baseline sharding protocols, for scenarios where bandwidths of miners range from 5Mbps to 60Mbps.

cs.CR↗

arXiv2024

MSfusion: A Dynamic Model Splitting Approach for Resource-Constrained Machines to Collaboratively Train Larger Models

Training large models requires a large amount of data, as well as abundant computation resources. While collaborative learning (e.g., federated learning) provides a promising paradigm to harness collective data from many participants, training large models remains a major challenge for participants with limited resources like mobile devices. We introduce MSfusion, an effective and efficient collaborative learning framework, tailored for training larger models on resourceconstraint machines through model splitting. Specifically, a double shifting model splitting scheme is designed such that in each training round, each participant is assigned a subset of model parameters to train over local data, and aggregates with sub-models of other peers on common parameters. While model splitting significantly reduces the computation and communication costs of individual participants, additional novel designs on adaptive model overlapping and contrastive loss functions help MSfusion to maintain training effectiveness, against model shift across participants. Extensive experiments on image and NLP tasks illustrate significant advantages of MSfusion in performance and efficiency for training large models, and its strong scalability: computation cost of each participant reduces significantly as the number of participants increases.

cs.LG↗

arXiv2024

BackdoorIndicator: Leveraging OOD Data for Proactive Backdoor Detection in Federated Learning

In a federated learning (FL) system, decentralized data owners (clients) could upload their locally trained models to a central server, to jointly train a global model. Malicious clients may plant backdoors into the global model through uploading poisoned local models, causing misclassification to a target class when encountering attacker-defined triggers. Existing backdoor defenses show inconsistent performance under different system and adversarial settings, especially when the malicious updates are made statistically close to the benign ones. In this paper, we first reveal the fact that planting subsequent backdoors with the same target label could significantly help to maintain the accuracy of previously planted backdoors, and then propose a novel proactive backdoor detection mechanism for FL named BackdoorIndicator, which has the server inject indicator tasks into the global model leveraging out-of-distribution (OOD) data, and then utilizing the fact that any backdoor samples are OOD samples with respect to benign samples, the server, who is completely agnostic of the potential backdoor types and target labels, can accurately detect the presence of backdoors in uploaded models, via evaluating the indicator tasks. We perform systematic and extensive empirical studies to demonstrate the consistently superior performance and practicality of BackdoorIndicator over baseline defenses, across a wide range of system and adversarial settings.

cs.CR↗

arXiv2024

Towards Client Driven Federated Learning

Conventional federated learning (FL) frameworks follow a server-driven model where the server determines session initiation and client participation, which faces challenges in accommodating clients' asynchronous needs for model updates. We introduce Client-Driven Federated Learning (CDFL), a novel FL framework that puts clients at the driving role. In CDFL, each client independently and asynchronously updates its model by uploading the locally trained model to the server and receiving a customized model tailored to its local task. The server maintains a repository of cluster models, iteratively refining them using received client models. Our framework accommodates complex dynamics in clients' data distributions, characterized by time-varying mixtures of cluster distributions, enabling rapid adaptation to new tasks with superior performance. In contrast to traditional clustered FL protocols that send multiple cluster models to a client to perform distribution estimation, we propose a paradigm that offloads the estimation task to the server and only sends a single model to a client, and novel strategies to improve estimation accuracy. We provide a theoretical analysis of CDFL's convergence. Extensive experiments across various datasets and system settings highlight CDFL's substantial advantages in model performance and computation efficiency over baselines.

cs.LG↗

arXiv2024

FedMeS: Personalized Federated Continual Learning Leveraging Local Memory

We focus on the problem of Personalized Federated Continual Learning (PFCL): a group of distributed clients, each with a sequence of local tasks on arbitrary data distributions, collaborate through a central server to train a personalized model at each client, with the model expected to achieve good performance on all local tasks. We propose a novel PFCL framework called Federated Memory Strengthening FedMeS to address the challenges of client drift and catastrophic forgetting. In FedMeS, each client stores samples from previous tasks using a small amount of local memory, and leverages this information to both 1) calibrate gradient updates in training process; and 2) perform KNN-based Gaussian inference to facilitate personalization. FedMeS is designed to be task-oblivious, such that the same inference process is applied to samples from all tasks to achieve good performance. FedMeS is analyzed theoretically and evaluated experimentally. It is shown to outperform all baselines in average accuracy and forgetting rate, over various combinations of datasets, task distributions, and client numbers.

cs.LG↗

arXiv2024

OmniLytics+: A Secure, Efficient, and Affordable Blockchain Data Market for Machine Learning through Off-Chain Processing

The rapid development of large machine learning (ML) models requires a massive amount of training data, resulting in booming demands of data sharing and trading through data markets. Traditional centralized data markets suffer from low level of security, and emerging decentralized platforms are faced with efficiency and privacy challenges. In this paper, we propose OmniLytics+, the first decentralized data market, built upon blockchain and smart contract technologies, to simultaneously achieve 1) data (resp., model) privacy for the data (resp. model) owner; 2) robustness against malicious data owners; 3) efficient data validation and aggregation. Specifically, adopting the zero-knowledge (ZK) rollup paradigm, OmniLytics+ proposes to secret share encrypted local gradients, computed from the encrypted global model, with a set of untrusted off-chain servers, who collaboratively generate a ZK proof on the validity of the gradient. In this way, the storage and processing overheads are securely offloaded from blockchain verifiers, significantly improving the privacy, efficiency, and affordability over existing rollup solutions. We implement the proposed OmniLytics+ data market as an Ethereum smart contract [41]. Extensive experiments demonstrate the effectiveness of OmniLytics+ in training large ML models in presence of malicious data owner, and the substantial advantages of OmniLytics+ in gas cost and execution time over baselines.

cs.CR↗

arXiv2023

Stochastic Coded Federated Learning: Theoretical Analysis and Incentive Mechanism Design

Federated learning (FL) has achieved great success as a privacy-preserving distributed training paradigm, where many edge devices collaboratively train a machine learning model by sharing the model updates instead of the raw data with a server. However, the heterogeneous computational and communication resources of edge devices give rise to stragglers that significantly decelerate the training process. To mitigate this issue, we propose a novel FL framework named stochastic coded federated learning (SCFL) that leverages coded computing techniques. In SCFL, before the training process starts, each edge device uploads a privacy-preserving coded dataset to the server, which is generated by adding Gaussian noise to the projected local dataset. During training, the server computes gradients on the global coded dataset to compensate for the missing model updates of the straggling devices. We design a gradient aggregation scheme to ensure that the aggregated model update is an unbiased estimate of the desired global update. Moreover, this aggregation scheme enables periodical model averaging to improve the training efficiency. We characterize the tradeoff between the convergence performance and privacy guarantee of SCFL. In particular, a more noisy coded dataset provides stronger privacy protection for edge devices but results in learning performance degradation. We further develop a contract-based incentive mechanism to coordinate such a conflict. The simulation results show that SCFL learns a better model within the given time and achieves a better privacy-performance tradeoff than the baseline methods. In addition, the proposed incentive mechanism grants better training performance than the conventional Stackelberg game approach.

cs.DC↗