Searcharxiv⌕ Search

arXiv subjects

Sicong Zhang

Publications and source records attributed to Sicong Zhang.

6 recordsLinked to original sources
arXiv2025

Sample-Independent Federated Learning Backdoor Attack in Speaker Recognition

In federated learning, backdoor attacks embed triggers in the adversarial client's data to inject a backdoor into the model. In order to enhance the stealth, an attack method based on the dropout layer has been proposed, which can implant the backdoor without modifying the sample. However, these methods struggle to covertly utilize dropout in evaluation mode, thus hindering their deployment in real-world scenarios. To address these, this paper introduces GhostB, a novel approach to federated learning backdoor attacks in speaker recognition that neither alters samples nor relies on dropout. This method employs the behavior of neurons producing specific values as triggers. By mapping these neuronal values to categories specified by the adversary, the backdoor is implanted and activated when particular feature values are detected at designated neurons. Our experiments conducted on TIMIT, LibriSpeech, and VoxCeleb2 databases in both Closed Set Identification (CSI) and Open Set Identification (OSI) scenarios demonstrate that GhostB achieves a 100% success rate upon activation in speaker recognition, with this rate maintained across experiments involving 1 to 50 ghost neurons. This paper investigates how the dispersion of neurons and their depth within hidden layers affect the success rate, revealing that increased dispersion and positioning of neurons can significantly decrease effectiveness, potentially rendering the attack unsuccessful.

cs.CR↗
arXiv2024

mLoRA: Fine-Tuning LoRA Adapters via Highly-Efficient Pipeline Parallelism in Multiple GPUs

Transformer-based, pre-trained large language models (LLMs) have demonstrated outstanding performance across diverse domains, particularly in the emerging {\em pretrain-then-finetune} paradigm. Low-Rank Adaptation (LoRA), a parameter-efficient fine-tuning method, is commonly used to adapt a base LLM to multiple downstream tasks. Further, LLM platforms enable developers to fine-tune multiple models and develop various domain-specific applications simultaneously. However, existing model parallelism schemes suffer from high communication overhead and inefficient GPU utilization when training multiple LoRA tasks across GPUs and machines. In this paper, we present mLoRA, a parallelism-efficient fine-tuning system designed for training multiple LoRA across GPUs and machines. mLoRA introduces a novel LoRA-aware pipeline parallelism scheme that efficiently pipelines independent LoRA adapters and their distinct fine-tuning stages across GPUs and machines, along with a new LoRA-efficient operator to enhance GPU utilization during pipelined LoRA training. Our extensive evaluation shows that mLoRA can significantly reduce average fine-tuning task completion time, e.g., by 30\%, compared to state-of-the-art methods like FSDP. More importantly, mLoRA enables simultaneous fine-tuning of larger models, e.g., two Llama-2-13B models on four NVIDIA RTX A6000 48GB GPUs, which is not feasible for FSDP due to high memory requirements. Hence, mLoRA not only increases fine-tuning efficiency but also makes it more accessible on cost-effective GPUs. mLoRA has been deployed in AntGroup's production environment.

cs.LG↗
arXiv2024

SAB:A Stealing and Robust Backdoor Attack based on Steganographic Algorithm against Federated Learning

Federated learning, an innovative network architecture designed to safeguard user privacy, is gaining widespread adoption in the realm of technology. However, given the existence of backdoor attacks in federated learning, exploring the security of federated learning is significance. Nevertheless, the backdoors investigated in current federated learning research can be readily detected by human inspection or resisted by detection algorithms. Accordingly, a new goal has been set to develop stealing and robust federated learning backdoor attacks. In this paper, we introduce a novel approach, SAB, tailored specifically for backdoor attacks in federated learning, presenting an alternative gradient updating mechanism. SAB attack based on steganographic algorithm, using image steganographic algorithm to build a full-size trigger to improve the accuracy of backdoors and use multiple loss joint computation to produce triggers. SAB exhibits smaller distances to benign samples and greater imperceptibility to the human eye. As such, our triggers are capable of mitigating or evading specific backdoor defense methods. In SAB, the bottom-95\% method is applied to extend the lifespan of backdoor attacks. It updates the gradient on minor value points to reduce the probability of being cleaned. Finally, the generalization of backdoors is enhanced with Sparse-update to improve the backdoor accuracy.

cs.CR↗
arXiv2014

Multi-crossing Number for Knots and the Kauffman Bracket Polynomial

A multi-crossing (or n-crossing) is a singular point in a projection at which n strands cross so that each strand bisects the crossing. We generalize the classic result of Kauffman, Murasugi, and Thistlethwaite, which gives the upper bound on the span of the bracket polynomial of K as 4c_2(K), to the n-crossing number: span is bounded above by ([n^2/2] + 4n-8) c_n(K) for all integers n at least 3. We also explore n-crossing additivity under composition, and find that for n at least 4, there are examples of knots such that the n-crossing number is sub-additive. Further, we present the first extensive list of calculations of n-crossing numbers for knots. Finally, we explore the monotonicity of the sequence of n-crossings of a knot, which we call the crossing spectrum.

math.GT↗
arXiv2013

Bounds on Übercrossing and Petal Numbers for Knots

An $n$-crossing is a point in the projection of a knot where $n$ strands cross so that each strand bisects the crossing. An übercrossing projection has a single $n$-crossing and a petal projection has a single $n$-crossing such that there are no loops nested within others. The übercrossing number, $\text{ü}(K)$, is the smallest $n$ for which we can represent a knot $K$ with a single $n$-crossing. The petal number is the number of loops in the minimal petal projection. In this paper, we relate the übercrossing number and petal number to well-known invariants such as crossing number, bridge number, and unknotting number. We find that the bounds we have constructed are tight for $(r, r+1)$-torus knots. We also explore the behavior of übercrossing number under composition.

math.GT↗