arXiv subjects

Toshiki Shibahara

Publications and source records attributed to Toshiki Shibahara.

11 recordsLinked to original sources

arXiv2026

Dummy Backdoor as a Defense: Removing Unknown Backdoors via Shared Internal Mechanisms for Generative LLMs

Backdoor attacks pose a serious threat to the safety and reliability of Large Language Models (LLMs), as they cause models to behave normally on clean inputs while producing attacker-specified responses when hidden triggers are present. Removing such unknown backdoors is particularly challenging when the defender does not know the backdoor attack types or the internal mechanisms formed through backdoor training. In this work, we propose a simple but effective backdoor removal method based on shared internal mechanisms across different backdoors. First, we show that different backdoors with the same task (attack objective) induce similar trigger-activated changes in the internal activations. Motivated by this observation, our method intentionally embeds a backdoor with a known trigger (\emph{dummy backdoor}) and then removes it through further fine-tuning on dummy-triggered inputs paired with clean responses. Since the dummy backdoor and the unknown backdoor can rely on shared internal mechanisms, removing the dummy backdoor also reduces the effect of the unknown backdoor. We evaluate our method on three backdoor attack types across multiple model families. Experimental results show that our method substantially reduces the attack success rate of the unknown backdoor while preserving model utility, outperforming representative existing defense methods in both backdoor removal effectiveness and utility preservation. These findings suggest that a defender-controllable backdoor can serve as a helpful proxy for mitigating unknown backdoors in generative LLMs.

cs.CR↗

arXiv2026

The Impact of Micro-level User Interventions on Macro-level Misinformation Spread

User interventions such as nudges, prebunking, and contextualization have been widely studied as countermeasures against misinformation, and shown to suppress individual users' sharing behavior. However, it remains unclear whether and to what extent such individual-level effects translate into reductions in collective misinformation prevalence. In this study, we incorporate user interventions as reductions in users' susceptibility within an empirically calibrated network-based misinformation diffusion model. We then systematically evaluate how intervention strength, scale, timing, target selection, and combinations of interventions affect overall misinformation prevalence through numerical simulations. The simulation results reveal that current user-level interventions may not necessarily produce sufficient collective effects. Specifically, each intervention alone only modestly suppresses misinformation prevalence, and even when design adjustments such as expanding intervention scale, implementing interventions earlier, or strategically selecting target users are introduced, the resulting gains in suppression remain limited. Although combining multiple interventions improves the suppression effect compared to using each intervention alone, achieving substantial reductions in misinformation prevalence remains difficult within realistically attainable intervention levels. This study quantitatively clarifies the gap between micro-level user interventions and macro-level misinformation spread, and demonstrates the limitations of evaluating misinformation countermeasures based solely on individual-level effectiveness.

cs.SI↗

arXiv2025

Robust Backdoor Removal by Reconstructing Trigger-Activated Changes in Latent Representation

Backdoor attacks pose a critical threat to machine learning models, causing them to behave normally on clean data but misclassify poisoned data into a poisoned class. Existing defenses often attempt to identify and remove backdoor neurons based on Trigger-Activated Changes (TAC) which is the activation differences between clean and poisoned data. These methods suffer from low precision in identifying true backdoor neurons due to inaccurate estimation of TAC values. In this work, we propose a novel backdoor removal method by accurately reconstructing TAC values in the latent representation. Specifically, we formulate the minimal perturbation that forces clean data to be classified into a specific class as a convex quadratic optimization problem, whose optimal solution serves as a surrogate for TAC. We then identify the poisoned class by detecting statistically small $L^2$ norms of perturbations and leverage the perturbation of the poisoned class in fine-tuning to remove backdoors. Experiments on CIFAR-10, GTSRB, and TinyImageNet demonstrated that our approach consistently achieves superior backdoor suppression with high clean accuracy across different attack types, datasets, and architectures, outperforming existing defense methods.

cs.LG↗

arXiv2025

Network Prebunking Problem: Optimizing Prebunking Targets to Suppress the Spread of Misinformation in Social Networks

As a countermeasure against misinformation that undermines the healthy use of social media, a preventive intervention known as \textit{prebunking} has recently attracted attention in the field of psychology. Prebunking aims to strengthen individuals' cognitive resistance to misinformation by presenting weakened doses of misinformation or by teaching common manipulation techniques before they encounter actual misinformation. Despite the growing body of evidence supporting its effectiveness in reducing susceptibility to misinformation at the individual level, an important open question remains: how best to identify the optimal targets for prebunking interventions to mitigate the spread of misinformation in a social network. To address this issue, we formulate a combinatorial optimization problem, called the \textit{network prebunking problem}, which aims to select optimal prebunking targets that minimizes the spread of misinformation in a social network under limited intervention budgets. We show that the problem is NP-hard and that its objective function is monotone and submodular, which provides a theoretical foundation for approximation guarantees of greedy algorithms. However, since the greedy algorithm is computationally expensive and does not scale to large networks, we propose an efficient approximation algorithm, MIA-NPP, based on the Maximum Influence Arborescence (MIA) approach, which restricts influence propagation around each node to a local directed tree rooted at that node. Through numerical experiments using real-world social network datasets, we demonstrate that MIA-NPP effectively suppresses the spread of misinformation under both fully observed and uncertain model parameter settings.

cs.SI↗

arXiv2025

Concept Unlearning in Large Language Models via Self-Constructed Knowledge Triplets

Machine Unlearning (MU) has recently attracted considerable attention as a solution to privacy and copyright issues in large language models (LLMs). Existing MU methods aim to remove specific target sentences from an LLM while minimizing damage to unrelated knowledge. However, these approaches require explicit target sentences and do not support removing broader concepts, such as persons or events. To address this limitation, we introduce Concept Unlearning (CU) as a new requirement for LLM unlearning. We leverage knowledge graphs to represent the LLM's internal knowledge and define CU as removing the forgetting target nodes and associated edges. This graph-based formulation enables a more intuitive unlearning and facilitates the design of more effective methods. We propose a novel method that prompts the LLM to generate knowledge triplets and explanatory sentences about the forgetting target and applies the unlearning process to these representations. Our approach enables more precise and comprehensive concept removal by aligning the unlearning process with the LLM's internal knowledge representations. Experiments on real-world and synthetic datasets demonstrate that our method effectively achieves concept-level unlearning while preserving unrelated knowledge.

cs.CL↗

arXiv2025

Sparse-Autoencoder-Guided Internal Representation Unlearning for Large Language Models

As large language models (LLMs) are increasingly deployed across various applications, privacy and copyright concerns have heightened the need for more effective LLM unlearning techniques. Many existing unlearning methods aim to suppress undesirable outputs through additional training (e.g., gradient ascent), which reduces the probability of generating such outputs. While such suppression-based approaches can control model outputs, they may not eliminate the underlying knowledge embedded in the model's internal activations; muting a response is not the same as forgetting it. Moreover, such suppression-based methods often suffer from model collapse. To address these issues, we propose a novel unlearning method that directly intervenes in the model's internal activations. In our formulation, forgetting is defined as a state in which the activation of a forgotten target is indistinguishable from that of ``unknown'' entities. Our method introduces an unlearning objective that modifies the activation of the target entity away from those of known entities and toward those of unknown entities in a sparse autoencoder latent space. By aligning the target's internal activation with those of unknown entities, we shift the model's recognition of the target entity from ``known'' to ``unknown'', achieving genuine forgetting while avoiding over-suppression and model collapse. Empirically, we show that our method effectively aligns the internal activations of the forgotten target, a result that the suppression-based approaches do not reliably achieve. Additionally, our method effectively reduces the model's recall of target knowledge in question-answering tasks without significant damage to the non-target knowledge.

cs.CL↗

arXiv2023

On Rényi Differential Privacy in Statistics-Based Synthetic Data Generation

Privacy protection with synthetic data generation often uses differentially private statistics and model parameters to quantitatively express theoretical security. However, these methods do not take into account privacy protection due to the randomness of data generation. In this paper, we theoretically evaluate Rényi differential privacy of the randomness in data generation of a synthetic data generation method that uses the mean vector and the covariance matrix of an original dataset. Specifically, for a fixed $α> 1$, we show the condition of $\varepsilon$ such that the synthetic data generation satisfies $(α, \varepsilon)$-Rényi differential privacy under a bounded neighboring condition and an unbounded neighboring condition, respectively. In particular, under the unbounded condition, when the size of the original dataset and synthetic datase is 10 million, the mechanism satisfies $(4, 0.576)$-Rényi differential privacy. We also show that when we translate it into the traditional $(\varepsilon, δ)$-differential privacy, the mechanism satisfies $(4.00, 10^{-10})$-differential privacy.

cs.CR↗

arXiv2023

Do Backdoors Assist Membership Inference Attacks?

When an adversary provides poison samples to a machine learning model, privacy leakage, such as membership inference attacks that infer whether a sample was included in the training of the model, becomes effective by moving the sample to an outlier. However, the attacks can be detected because inference accuracy deteriorates due to poison samples. In this paper, we discuss a \textit{backdoor-assisted membership inference attack}, a novel membership inference attack based on backdoors that return the adversary's expected output for a triggered sample. We found three crucial insights through experiments with an academic benchmark dataset. We first demonstrate that the backdoor-assisted membership inference attack is unsuccessful. Second, when we analyzed loss distributions to understand the reason for the unsuccessful results, we found that backdoors cannot separate loss distributions of training and non-training samples. In other words, backdoors cannot affect the distribution of clean samples. Third, we also show that poison and triggered samples activate neurons of different distributions. Specifically, backdoors make any clean sample an inlier, contrary to poisoning samples. As a result, we confirm that backdoors cannot assist membership inference.

cs.CR↗

arXiv2022

Interpreting Graph-based Sybil Detection Methods as Low-Pass Filtering

Online social networks (OSNs) are threatened by Sybil attacks, which create fake accounts (also called Sybils) on OSNs and use them for various malicious activities. Therefore, Sybil detection is a fundamental task for OSN security. Most existing Sybil detection methods are based on the graph structure of OSNs, and various methods have been proposed recently. However, although almost all methods have been compared experimentally in terms of detection performance and noise robustness, theoretical understanding of them is still lacking. In this study, we show that existing graph-based Sybil detection methods can be interpreted in a unified framework of low-pass filtering. This framework enables us to theoretically compare and analyze each method from two perspectives: filter kernel properties and the spectrum of shift matrices. Our analysis reveals that the detection performance of each method depends on how well low-pass filtering can extract low frequency components and remove noisy high frequency components. Furthermore, on the basis of the analysis, we propose a novel Sybil detection method called SybilHeat. Numerical experiments on synthetic graphs and real social networks demonstrate that SybilHeat performs consistently well on graphs with various structural properties. This study lays a theoretical foundation for graph-based Sybil detection and leads to a better understanding of Sybil detection methods.

cs.CR↗

arXiv2021

MEGEX: Data-Free Model Extraction Attack against Gradient-Based Explainable AI

The advance of explainable artificial intelligence, which provides reasons for its predictions, is expected to accelerate the use of deep neural networks in the real world like Machine Learning as a Service (MLaaS) that returns predictions on queried data with the trained model. Deep neural networks deployed in MLaaS face the threat of model extraction attacks. A model extraction attack is an attack to violate intellectual property and privacy in which an adversary steals trained models in a cloud using only their predictions. In particular, a data-free model extraction attack has been proposed recently and is more critical. In this attack, an adversary uses a generative model instead of preparing input data. The feasibility of this attack, however, needs to be studied since it requires more queries than that with surrogate datasets. In this paper, we propose MEGEX, a data-free model extraction attack against a gradient-based explainable AI. In this method, an adversary uses the explanations to train the generative model and reduces the number of queries to steal the model. Our experiments show that our proposed method reconstructs high-accuracy models -- 0.97$\times$ and 0.98$\times$ the victim model accuracy on SVHN and CIFAR-10 datasets given 2M and 20M queries, respectively. This implies that there is a trade-off between the interpretability of models and the difficulty of stealing them.

cs.CR↗

arXiv2017

A Study on the Vulnerabilities of Mobile Apps associated with Software Modules

This paper reports a large-scale study that aims to understand how mobile application (app) vulnerabilities are associated with software libraries. We analyze both free and paid apps. Studying paid apps was quite meaningful because it helped us understand how differences in app development/maintenance affect the vulnerabilities associated with libraries. We analyzed 30k free and paid apps collected from the official Android marketplace. Our extensive analyses revealed that approximately 70%/50% of vulnerabilities of free/paid apps stem from software libraries, particularly from third-party libraries. Somewhat paradoxically, we found that more expensive/popular paid apps tend to have more vulnerabilities. This comes from the fact that more expensive/popular paid apps tend to have more functionality, i.e., more code and libraries, which increases the probability of vulnerabilities. Based on our findings, we provide suggestions to stakeholders of mobile app distribution ecosystems.

cs.CR↗