arXiv subjects

Tomasz Michalak

Publications and source records attributed to Tomasz Michalak.

13 recordsLinked to original sources

arXiv2025

Adversarial Robustness of Link Sign Prediction in Signed Graphs

Signed graphs serve as fundamental data structures for representing positive and negative relationships in social networks, with signed graph neural networks (SGNNs) emerging as the primary tool for their analysis. Our investigation reveals that balance theory, while essential for modeling signed relationships in SGNNs, inadvertently introduces exploitable vulnerabilities to black-box attacks. To showcase this, we propose balance-attack, a novel adversarial strategy specifically designed to compromise graph balance degree, and develop an efficient heuristic algorithm to solve the associated NP-hard optimization problem. While existing approaches attempt to restore attacked graphs through balance learning techniques, they face a critical challenge we term "Irreversibility of Balance-related Information," as restored edges fail to align with original attack targets. To address this limitation, we introduce Balance Augmented-Signed Graph Contrastive Learning (BA-SGCL), an innovative framework that combines contrastive learning with balance augmentation techniques to achieve robust graph representations. By maintaining high balance degree in the latent space, BA-SGCL not only effectively circumvents the irreversibility challenge but also significantly enhances model resilience. Extensive experiments across multiple SGNN architectures and real-world datasets demonstrate both the effectiveness of our proposed balance-attack and the superior robustness of BA-SGCL, advancing the security and reliability of signed graph analysis in social networks. Datasets and codes of the proposed framework are at the github repository https://anonymous.4open.science/r/BA-SGCL-submit-DF41/.

cs.LG↗

arXiv2025

Learning Graph Representation of Agent Diffusers

Diffusion-based generative models have significantly advanced text-to-image synthesis, demonstrating impressive text comprehension and zero-shot generalization. These models refine images from random noise based on textual prompts, with initial reliance on text input shifting towards enhanced visual fidelity over time. This transition suggests that static model parameters might not optimally address the distinct phases of generation. We introduce LGR-AD (Learning Graph Representation of Agent Diffusers), a novel multi-agent system designed to improve adaptability in dynamic computer vision tasks. LGR-AD models the generation process as a distributed system of interacting agents, each representing an expert sub-model. These agents dynamically adapt to varying conditions and collaborate through a graph neural network that encodes their relationships and performance metrics. Our approach employs a coordination mechanism based on top-$k$ maximum spanning trees, optimizing the generation process. Each agent's decision-making is guided by a meta-model that minimizes a novel loss function, balancing accuracy and diversity. Theoretical analysis and extensive empirical evaluations show that LGR-AD outperforms traditional diffusion models across various benchmarks, highlighting its potential for scalable and flexible solutions in complex image generation tasks. Code is available at: https://github.com/YousIA/LGR_AD

cs.LG↗

arXiv2024

Bribe & Fork: Cheap Bribing Attacks via Forking Threat

In this work, we reexamine the vulnerability of Payment Channel Networks (PCNs) to bribing attacks, where an adversary incentivizes blockchain miners to deliberately ignore a specific transaction to undermine the punishment mechanism of PCNs. While previous studies have posited a prohibitive cost for such attacks, we show that this cost may be dramatically reduced (to approximately \$125), thereby increasing the likelihood of these attacks. To this end, we introduce Bribe & Fork, a modified bribing attack that leverages the threat of a so-called feather fork which we analyze with a novel formal model for the mining game with forking. We empirically analyze historical data of some real-world blockchain implementations to evaluate the scale of this cost reduction. Our findings shed more light on the potential vulnerability of PCNs and highlight the need for robust solutions.

cs.CR↗

arXiv2023

Hiders' Game

Consider spies infiltrating a network or dissidents secretly organising under a dictatorship. Such scenarios can be cast as adversarial social network analysis problems involving nodes connecting while evading network analysis tools, e.g., centrality measures or community detection algorithms. While most works consider unilateral actions of an evader, we define a network formation game. Here, several newcomers attempt to rewire the existing social network, to become directly tied with the high centrality players, while keeping their own centrality small. This extends the network formation literature, including the Jackson and Wolinsky model, by considering additional strategies and new utility functions. We algorithmically demonstrate that the pairwise Nash stable networks (\PANS) constitute a lattice, where the stronger \PANS{} lattice is nested in the weaker \PANS. We also prove that inclusion in \PANS{} implies less utility for everyone. Furthermore, we bound the social efficiency of \PANS{} and directly connect efficiency to the strength of \PANS. Finally, we characterise the \PANS{} in practically important settings, deriving tight efficiency bounds. Our results suggest the hiders how to interconnect stably and efficiently. Additionally, the results let us detect infiltrated networks, enhancing the social network analysis tools. Besides the theoretical development, this is applicable to fighting terrorism and espionage.

cs.GT↗

arXiv2023

Towards Secrecy-Aware Attacks Against Trust Prediction in Signed Social Networks

Signed social networks are widely used to model the trust relationships among online users in security-sensitive systems such as cryptocurrency trading platforms, where trust prediction plays a critical role. In this paper, we investigate how attackers could mislead trust prediction by secretly manipulating signed networks. To this end, we first design effective poisoning attacks against representative trust prediction models. The attacks are formulated as hard bi-level optimization problems, for which we propose several efficient approximation solutions. However, the resulting basic attacks would severely change the structural semantics (in particular, both local and global balance properties) of a signed network, which makes the attacks prone to be detected by the powerful attack detectors we designed. Given this, we further refine the basic attacks by integrating some conflicting metrics as penalty terms into the objective function. The refined attacks become secrecy-aware, i.e., they can successfully evade attack detectors with high probability while sacrificing little attack performance. We conduct comprehensive experiments to demonstrate that the basic attacks can severely disrupt trust prediction but could be easily detected, and the refined attacks perform almost equally well while evading detection. Overall, our results significantly advance the knowledge in designing more practical attacks, reflecting more realistic threats to current trust prediction models. Moreover, the results also provide valuable insights and guidance for building up robust trust prediction systems.

cs.CR↗

arXiv2023

Lightning Creation Games

Payment channel networks (PCNs) are a promising solution to the scalability problem of cryptocurrencies. Any two users connected by a payment channel in the network can theoretically send an unbounded number of instant, costless transactions between them. Users who are not directly connected can also transact with each other in a multi-hop fashion. In this work, we study the incentive structure behind the creation of payment channel networks, particularly from the point of view of a single user that wants to join the network. We define a utility function for a new user in terms of expected revenue, expected fees, and the cost of creating channels, and then provide constant factor approximation algorithms that optimise the utility function given a certain budget. Additionally, we take a step back from a single user to the whole network and examine the parameter spaces under which simple graph topologies form a Nash equilibrium.

cs.GT↗

arXiv2023

On Manipulating Weight Predictions in Signed Weighted Networks

Adversarial social network analysis studies how graphs can be rewired or otherwise manipulated to evade social network analysis tools. While there is ample literature on manipulating simple networks, more sophisticated network types are much less understood in this respect. In this paper, we focus on the problem of evading FGA -- an edge weight prediction method for signed weighted networks by Kumar et al.. Among others, this method can be used for trust prediction in reputation systems. We study the theoretical underpinnings of FGA and its computational properties in terms of manipulability. Our positive finding is that, unlike many other tools, this measure is not only difficult to manipulate optimally, but also it can be difficult to manipulate in practice.

cs.SI↗

arXiv2021

Multiagent Model-based Credit Assignment for Continuous Control

Deep reinforcement learning (RL) has recently shown great promise in robotic continuous control tasks. Nevertheless, prior research in this vein center around the centralized learning setting that largely relies on the communication availability among all the components of a robot. However, agents in the real world often operate in a decentralised fashion without communication due to latency requirements, limited power budgets and safety concerns. By formulating robotic components as a system of decentralised agents, this work presents a decentralised multiagent reinforcement learning framework for continuous control. To this end, we first develop a cooperative multiagent PPO framework that allows for centralized optimisation during training and decentralised operation during execution. However, the system only receives a global reward signal which is not attributed towards each agent. To address this challenge, we further propose a generic game-theoretic credit assignment framework which computes agent-specific reward signals. Last but not least, we also incorporate a model-based RL module into our credit assignment framework, which leads to significant improvement in sample efficiency. We demonstrate the effectiveness of our framework on experimental results on Mujoco locomotion control tasks. For a demo video please visit: https://youtu.be/gFyVPm4svEY.

cs.AI↗

arXiv2021

Strategic Evasion of Centrality Measures

Among the most fundamental tools for social network analysis are centrality measures, which quantify the importance of every node in the network. This centrality analysis typically disregards the possibility that the network may have been deliberately manipulated to mislead the analysis. To solve this problem, a recent study attempted to understand how a member of a social network could rewire the connections therein to avoid being identified as a leader of that network. However, the study was based on the assumption that the network analyzer - the seeker - is oblivious to any evasion attempts by the evader. In this paper, we relax this assumption by modelling the seeker and evader as strategic players in a Bayesian Stackelberg game. In this context, we study the complexity of various optimization problems, and analyze the equilibria of the game under different assumptions, thereby drawing the first conclusions in the literature regarding which centralities the seeker should use to maximize the chances of detecting a strategic evader.

cs.SI↗

arXiv2019

A Game-Theoretic Algorithm for Link Prediction

Predicting edges in networks is a key problem in social network analysis and involves reasoning about the relationships between nodes based on the structural properties of a network. In particular, link prediction can be used to analyse how a network will develop or - given incomplete information about relationships - to discover "missing" links. Our approach to this problem is rooted in cooperative game theory, where we propose a new, quasi-local approach (i.e., one which considers nodes within some radius k) that combines generalised group closeness centrality and semivalue interaction indices. We develop fast algorithms for computing our measure and evaluate it on a number of real-world networks, where it outperforms a selection of other state-of-the-art methods from the literature. Importantly, choosing the optimal radius k for quasi-local methods is difficult, and there is no assurance that the choice is optimal. Additionally, when compared to other quasi-local methods, ours achieves very good results even when given a suboptimal radius k as a parameter.

cs.SI↗

arXiv2016

Hiding Individuals and Communities in a Social Network

The Internet and social media have fueled enormous interest in social network analysis. New tools continue to be developed and used to analyse our personal connections, with particular emphasis on detecting communities or identifying key individuals in a social network. This raises privacy concerns that are likely to exacerbate in the future. With this in mind, we ask the question: Can individuals or groups actively manage their connections to evade social network analysis tools? By addressing this question, the general public may better protect their privacy, oppressed activist groups may better conceal their existence, and security agencies may better understand how terrorists escape detection. We first study how an individual can evade "network centrality" analysis without compromising his or her influence within the network. We prove that an optimal solution to this problem is hard to compute. Despite this hardness, we demonstrate that even a simple heuristic, whereby attention is restricted to the individual's immediate neighbourhood, can be surprisingly effective in practice. For instance, it could disguise Mohamed Atta's leading position within the WTC terrorist network, and that is by rewiring a strikingly-small number of connections. Next, we study how a community can increase the likelihood of being overlooked by community-detection algorithms. We propose a measure of concealment, expressing how well a community is hidden, and use it to demonstrate the effectiveness of a simple heuristic, whereby members of the community either "unfriend" certain other members, or "befriend" some non-members, in a coordinated effort to camouflage their community.

cs.SI↗

arXiv2014

A Measure of Synergy in Coalitions

When the performance of a team of agents exceeds our expectations or fall short of them, we often explain this by saying that there was some synergy in the team---either positive (the team exceeded our expectations) or negative (they fell short). Our aim in this article is to develop a formal and principled way of measuring synergies, both positive and negative. Using characteristic function cooperative games as our underlying model, we present a formal measure of synergy, based on the idea that a synergy is exhibited when the performance of a team deviates from the norm. We then show that our synergy value is the only possible such measure that satisfies certain intuitive properties. We then investigate some alternative characterisations of this measure.

cs.GT↗

arXiv2014

Towards a Fair Allocation of Rewards in Multi-Level Marketing

An increasing number of businesses and organisations rely on existing users for finding new users or spreading a message. One of the widely used "refer-a-friend" mechanisms offers an equal reward to both the referrer and the invitee. This mechanism provides incentives for direct referrals and is fair to the invitee. On the other hand, multi-level marketing and recent social mobilisation experiments focus on mechanisms that incentivise both direct and indirect referrals. Such mechanisms share the reward for inviting a new member among the ancestors, usually in geometrically decreasing shares. A new member receives nothing at the time of joining. We study fairness in multi-level marketing mechanisms. We show how characteristic function games can be used to model referral marketing, show how the canonical fairness concept of the Shapley value can be applied to this setting, and establish the complexity of finding the Shapley value in each class, and provide a comparison of the Shapley value-based mechanism to existing referral mechanisms.

cs.GT↗