The deployment of machine learning models in operational contexts represents a significant investment for any organisation. Consequently, the risk of these models being misappropriated by competitors needs to be addressed. In recent years, numerous proposals have been put forth to detect instances of model stealing. However, these proposals operate under implicit and disparate data and model access assumptions; as a consequence, it remains unclear how they can be effectively compared to one another. Our evaluation shows that a simple baseline that we introduce performs on par with existing state-of-the-art fingerprints, which, on the other hand, are much more complex. To uncover the reasons behind this intriguing result, this paper introduces a systematic approach to both the creation of model fingerprinting schemes and their evaluation benchmarks. By dividing model fingerprinting into three core components – Query, Representation and Detection (QuRD) – we are able to identify ~100 previously unexplored QuRD combinations and gain insights into their performance. Finally, we introduce a set of metrics to compare and guide the creation of more representative model stealing detection benchmarks. Our approach reveals the need for more challenging benchmarks and a sound comparison with baselines. To foster the creation of new fingerprinting schemes and benchmarks, we open-source our fingerprinting toolbox.

Among the many technical challenges to enforcing AI regulations, one crucial yet underexplored problem is the risk of audit manipulation. This manipulation occurs when a platform deliberately alters its answers to a regulator to pass an audit without modifying its answers to other users. In this paper, we introduce a novel approach to manipulation-proof auditing by taking into account the auditor's prior knowledge of the task solved by the platform. We first demonstrate that regulators must not rely on public priors (e.g. a public dataset), as platforms could easily fool the auditor in such cases. We then formally establish the conditions under which an auditor can prevent audit manipulations using prior knowledge about the ground truth. Finally, our experiments with two standard datasets illustrate the maximum level of unfairness a platform can hide before being detected as malicious. Our formalization and generalization of manipulation-proof auditing with a prior opens up new research directions for more robust fairness audits.

Existing work in fairness auditing assumes that each audit is performed independently. In this paper, we consider multiple agents working together, each auditing the same platform for different tasks. Agents have two levers: their collaboration strategy, with or without coordination beforehand, and their strategy for sampling appropriate data points. We theoretically compare the interplay of these levers. Our main findings are that (i) collaboration is generally beneficial for accurate audits, (ii) basic sampling methods often prove to be effective, and (iii) counter-intuitively, extensive coordination on queries often deteriorates audits accuracy as the number of agents increases. Experiments on three large datasets confirm our theoretical results. Our findings motivate collaboration during fairness audits of platforms that use ML models for decision-making.

Modern communication networks feature fully decen-tralized flow rerouting mechanisms which allow them to quickly react to link failures. This paper revisits the fundamental algorithmic problem underlying such local fast rerouting mechanisms. Is it possible to achieve perfect resilience, i.e., to define local routing tables which preserve connectivity as long as the underlying network is still connected? Feigenbaum et al. [1] and Foerster et al. [2] showed that, unfortunately, it is impossible in general.This paper charts a more complete landscape of the feasibility of perfect resilience. We first show a perhaps surprisingly large price of locality in static fast rerouting mechanisms: even when source and destination remain connected by a linear number of link-disjoint paths after link failures, local rerouting algorithms cannot find any of them which leads to a disconnection on the routing level. This motivates us to study resilience in graphs which exclude certain dense minors, such as cliques or a complete bipartite graphs, and in particular, provide characterizations of the possibility of perfect resilience in different routing models. We provide further insights into the price of locality by showing impossibility results for few failures and investigate perfect resilience on Topology Zoo networks.

To provide fast traffic recovery upon failures, most modern networks support static Fast Rerouting (FRR) mechanisms for mission critical services. However, configuring FRR mechanisms to toleratemultiplefailures poses challenging algorithmic problems. While state-of-the-art solutions leveraging arc-disjoint arborescence-based network decompositions ensure that failover routes always reach their destinations eventually, even under multiple concurrent failures, these routes may be long and introduce unnecessary loads; moreover, they are tailored to worst-case failure scenarios. This article presents an algorithmic framework for improving a given FRR network decomposition,using postprocessing. In particular, our framework is based on iterative arc swapping strategies and supports a number of use cases, from strengthening the resilience (e.g., in the presence of shared risk link groups) to improving the quality of the resulting routes (e.g., reducing route lengths and induced loads). Our simulations show that postprocessing is indeed beneficial in various scenarios, and can therefore enhance today’s approaches.

To provide a high availability and to be able to quickly react to link failures, most communication networks feature fast rerouting (FRR) mechanisms in the data plane. However, configuring these mechanisms to provide a high resilience against multiple failures is algorithmically challenging, as rerouting rules can only depend on local failure information and need to be pre-defined. This paper is motivated by the observation that the common approach to design fast rerouting algorithms, based on spanning trees and covering arborescences, comes at a cost of reduced resilience as it does not fully exploit the available links in heterogeneous topologies. We present several novel fast rerouting algorithms which are not limited by spanning trees, but rather extend and combine ("graft") multiple spanning arborescences to improve resilience. We compare our algorithms analytically and empirically, and show that they can significantly improve not only the resilience, but also accelerate the preprocessing to generate the local fast failover rules.

Shadow banning consists for an online social net-work in limiting the visibility of some of its users, without them being aware of it. Twitter declares that it does not use such a practice, sometimes arguing about the occurrence of "bugs" to justify restrictions on some users. This paper is the first to address the plausibility of shadow banning on a major online platform, by adopting both a statistical and a graph topological approach.We first conduct an extensive data collection and analysis campaign, gathering occurrences of visibility limitations on user profiles (we crawl more than 2.5 millions of them). In such a black-box observation setup, we highlight the salient user profile features that may explain a banning practice (using machine learning predictors). We then pose two hypotheses for the phenomenon: i) limitations are bugs, as claimed by Twitter, and ii) shadow banning propagates as an epidemic on user-interaction ego-graphs. We show that hypothesis i) is statistically unlikely with regards to the data we collected. We then show some interesting correlation with hypothesis ii), suggesting that the interaction topology is a good indicator of the presence of groups of shadow banned users on the service.

In network optimization problems, from traffic engineering to network monitoring, the routing model is typically considered as something given and frozen. This paper is motivated by the fundamental question how the ability tochangeandoptimizethe routing model itself influences the efficiency at which communication networks can be operated. To this end, we identify two main dimensions of a routing model:consistency(of a single route) andcoherence(of sets of routes). We present analytical results on the impact of the routing model on the achievable route diversity as well as on the runtime of solving optimization problems underlying different case studies. We also uncover that it can sometimes be beneficial toartificiallyrestrict the routing model, to significantly reduce the computational complexity without negatively affecting the route diversity much.

In order to provide a high resilience and to react quickly to link failures, modern computer networks support fully decentralized flow rerouting, also known as local fast failover. In a nutshell, the task of a local fast failover algorithm is to pre-define fast failover rules for each node using locally available information only. Ideally, such a local fast failover algorithm provides a perfect resilience deterministically: a packet emitted from any source can reach any target, as long as the underlying network remains connected. Feigenbaum et al. showed [Feigenbaum and others, 2012] that it is not always possible to provide perfect resilience; on the positive side, the authors also presented an efficient algorithm which achieves at least 1-resilience, tolerating a single failure in any network. Interestingly, not much more is known currently about the feasibility of perfect resilience. This brief announcement revisits perfect resilience with local fast failover, both in a model where the source can and cannot be used for forwarding decisions. By establishing a connection between graph minors and resilience, we prove that it is impossible to achieve perfect resilience on any non-planar graph; On the positive side, we can derive perfect resilience for outerplanar and some planar graphs.

To provide high availability despite link failures, many modern communication networks feature fast failover mechanisms in the data plane, which operates orders of magnitude faster than the control plane. While the configuration of highly resilient data planes is known to be a difficult combinatorial problem, over the last years, much progress has been made in the design of algorithms which provably guarantee connectivity even under many concurrent link failures. However, while these algorithms provide connectivity, the resulting routes after failures can be very long, which in turn can harm performance. In this paper, we propose, analyze, and evaluate methods for fast failover algorithms which account for the quality of the routes after failures, in addition to connectivity. In particular, we revisit the existing approach to cover the to-be-protected network with arc-disjoint spanning arborescences to define alternative routes to the destination, aiming to keep the stretch imposed by these trees low (hence the name of our method: Bonsai). We show that the underlying problem is NP-hard on general topologies and present lower bound results that are tight for various topologies, for any class of fast failover algorithms. We also present heuristics for general networks and demonstrate their performance benefits in extensive simulations. Finally, we show that failover algorithms using low-stretch arborescences, as a side effect, can provide connectivity under more general failure models than usually considered in the literature.

To meet the stringent requirements on the maximally tolerable disruptions of traffic under link failures, many communication networks feature some sort of static failover mechanism for fast rerouting. However, configuring such static failover mechanisms to achieve a high degree of robustness is known to be challenging, in particular when packet tagging or dynamic node state cannot be used. This paper initiates the systematic study of such local fast failover mechanisms which not only provide connectivity guarantees, even under multiple link failures, but also account for the quality of the resulting failover routes, with respect to locality (i.e., route length) and congestion. Failover quality has received less attention in the literature so far, yet it is increasingly important to support emerging applications.We first show that there exists an inherent tradeoff in terms of achievable locality and congestion of failover routes. We then present CASA, an algorithm providing a high degree of robustness as well as a provable quality of fast rerouting. CASA combines two crucial static resilient routing techniques: combinatorial designs and arc-disjoint arborescences. We complement our formal analysis with a simulation study, in which we compare our algorithms with the state-of-the-art in different scenarios and show benefits in terms of stretch, load, and resilience.

Neural networks are powering the deployment of embedded devices and Internet of Things. Applications range from personal assistants to critical ones such as self-driving cars. It has been shown recently that models obtained from neural nets can be trojaned; an attacker can then trigger an arbitrary model behavior facing crafted inputs. This has a critical impact on the security and reliability of those deployed devices. We introduce novel algorithms to detect the tampering with deployed models, classifiers in particular. In the remote interaction setup we consider, the proposed strategy is to identify markers of the model input space that are likely to change class if the model is attacked, allowing a user to detect a possible tampering. This setup makes our proposal compatible with a wide range of scenarios, such as embedded models, or models exposed through prediction APIs. We experiment those tampering detection algorithms on the canonical MNIST dataset, over three different types of neural nets, and facing five different attacks (trojaning, quantization, fine-tuning, compression and watermarking). We then validate over five large models (VGG16, VGG19, ResNet, MobileNet, DenseNet) with a state of the art dataset (VGGFace2), and report results demonstrating the possibility of an efficient detection of model tampering.

Modern online applications value real-time queries over fresh data models. This is the case for graph-based applications, such as social networking or recommender systems, running on front-end servers in production. A core problem in graph processing systems is the efficient partitioning of the input graph over multiple workers. Recent advances over Bulk Synchronous Parallel processing systems (BSP) enabled computations over partitions on those workers, independently of global synchronization supersteps. A good objective partitioning makes the understanding of the load balancing and communication trade-off mandatory for performance improvement. This short paper addresses this trade-off through the proposal of an optimization problem, that is to be solved continuously to avoid performance degradation over time. Our simulations show that the design of the software module we propose yields significant performance improvements over the BSP processing model.

To provide fast traffic recovery upon failures, most modern networks support static Fast Rerouting (FRR) mechanisms for mission critical services. However, configuring FRR mechanisms to tolerate multiple failures poses challenging algorithmic problems. While state-of-the-art solutions leveraging arc-disjoint arborescence-based network decompositions ensure that failover routes always reach their destinations eventually, even under multiple concurrent failures, these routes may be long and introduce unnecessary loads; moreover, they are tailored to worst-case failure scenarios. This paper presents an algorithmic framework for improving a given FRR network decomposition, using postprocessing. In particular, our framework is based on iterative arc swapping strategies and supports a number of use cases, from strengthening the resilience (e.g., in the presence of shared risk link groups) to improving the quality of the resulting routes (e.g., reducing route lengths and induced loads). Our simulations show that postprocessing is indeed beneficial in various scenarios, and can therefore enhance today's approaches.

Reliable and highly available computer networks must implement resilient fast rerouting mechanisms: upon a link or node failure, an alternative route is determined quickly, without involving the network control plane. Designing such fast failover mechanisms capable of dealing with multiple concurrent failures, however, is challenging, as failover rules need to be installed proactively, i.e., ahead of time, without knowledge of the actual failures happening at runtime. Indeed, only little is known today about the design of resilient routing algorithms. This paper introduces a general framework to reason about and design local failover algorithms that minimize the resulting load after failover on dense networks, beyond destination-based routing. We show that due to the inherent locality of the failover decisions at runtime, the problem is fundamentally related to the field of distributed algorithms without coordination. We derive an intriguing lower bound on the inherent network load overhead any local fast failover scheme that will introduce in the worst case, even though globally seen, much more balanced traffic allocations exist. We then present different randomized and deterministic failover algorithms and analyze their overhead load. In particular, we build upon the theory of combinatorial designs and develop a novel deterministic failover mechanism based on symmetric block design theory, which tolerates a maximal number of link failures while ensuring low loads.

Ensuring the early detection of important social network users is a challenging task. Some peer ranking services are now well established, such as PeerIndex, Klout, or Kred. Their function is to rank users according to their influence. This notion of influence is however abstract, and the algorithms achieving this ranking are opaque. Following the rising demand for a more transparent web, we explore the problem of gaining knowledge by reverse engineering such peer ranking services, with regards to the social network topology they get as an input. Since these services exploit the online activity of users (and therefore their connectivity in social networks), we provide a precise evaluation of how topological metrics of the social network impact the final user ranking. Our approach is the following: we first model the ranking service as a black-box with which we interact by creating user profiles and by performing operations on them. Through those profiles, we trigger some slight topological modifications. By monitoring the impact of these modifications on the rankings of those profiles, we infer the weight of each topological metric in the black-box, thus reversing the service influence cookbook.

Reliable and highly available computer networks must implement resilient fast rerouting mechanisms: upon a link or node failure, an alternative route is determined quickly, without involving the network control plane. Designing such fast failover mechanisms capable of dealing with multiple concurrent failures however is challenging, as failover rules need to be installed proactively, i.e., ahead of time, without knowledge of the actual failures happening at runtime. Indeed, only little is known today about the design of resilient routing algorithms. This paper presents a deterministic local failover mechanism which we prove to result in a minimum network load for a wide range of communication patterns, solving an open problem. Our mechanism relies on the key insight that resilient routing essentially constitutes a distributed algorithm without coordination. Accordingly, we build upon the theory of combinatorial designs and develop a novel deterministic failover mechanism based on symmetric block design theory which tolerates a maximal number of Ω(n) link failures in an n-node network and in the worst-case, while always ensuring routing connectivity. In particular, we show that at least Ω(φ2) link failures are needed to generate a maximum link load of at least φ, which matches an existing bound on the number of link failures needed for an optimal failover scheme. We complement our formal analysis with simulations, showing that our approach outperforms prior schemes not only in the worst-case.

Event logging is a key source of information on a system state. Reading logs provides insights on its activity, assess its correct state and allows to diagnose problems. However, reading does not scale: with the number of machines increasingly rising, and the complexification of systems, the task of auditing systems' health based on logfiles is becoming overwhelming for system administrators. This observation led to many proposals automating the processing of logs. However, most of these proposal still require some human intervention, for instance by tagging logs, parsing the source files generating the logs, etc. In this work, we target minimal human intervention for logfile processing and propose a new approach that considers logs as regular text (as opposed to related works that seek to exploit at best the little structure imposed by log formatting). This approach allows to leverage modern techniques from natural language processing. More specifically, we first apply a word embedding technique based on Google's word2vec algorithm: logfiles' words are mapped to a high dimensional metric space, that we then exploit as a feature space using standard classifiers. The resulting pipeline is very generic, computationally efficient, and requires very little intervention. We validate our approach by seeking stress patterns on an experimental platform. Results show a strong predictive performance (≈ 90% accuracy) using three out-of-the-box classifiers.

Recent studies have introduced co-location attacks as a powerful way to extract social information from location traces. However, these attacks all rely by some means on the position of targeted users. This requires the attacker to be able to locate either the user or the sensors detecting the user. Implicitly, it also forbids the use of these attacks on devices whose location is unknown.

This paper presents a methodology to guide the evaluation of social distributed applications in mobile environments. Even when applications are already designed, they exhibit a number of tuning parameters upon which network operators can act in order to improve performance. Accordingly, evaluation can be a valuable tool to determine for a particular mobile application which is the most suitable parameters setup from a performance point of view. Our methodology can be of great interest in this tuning process, thus saving both time and money. The main novelty of this methodology is the use of diversification to recreate mobile environments using both synthetic and real mobility traces. Our work focuses on how micro-mobility may impact social distributed applications. The feasibility of the paper is showed through a realistic microblogging case study.

In many large‐scale distributed systems and on the Web, agents need to interact with other unknown agents to carry out some tasks or transactions. The ability to reason about and assess the potential risks in carrying out such transactions is essential for providing a safe and reliable interaction environment. A traditional approach to reason about the risk of a transaction is to determine if the involved agent is trustworthy on the basis of its behavior history. As a departure from such traditional trust models, we propose a generic, trust framework based on machine learning where an agent uses its own previous transactions (with other agents) to build a personal knowledge base. This is used to assess the trustworthiness of a transaction on the basis of the associated features, particularly using the features that help discern successful transactions from unsuccessful ones. These features are handled by applying appropriate machine learning algorithms to extract the relationships between the potential transaction and the previous ones. Experiments based on real data sets show that our approach is more accurate than other trust mechanisms, especially when the information about past behavior of the specific agent is rare, incomplete, or inaccurate.

Simulating human-centered pervasive systems requires accurate assumptions on the behavior of human groups. Recent models consider this behavior as a combination of both social and spatial factors. Yet, establishing accurate traces of human groups is difficult: current techniques capture either positions, or contacts, with a limited accuracy.

This paper demonstrates that virtual networks that are dynamically embedded on a given resource network may constitute a security threat as properties of the infrastructure-typically a business secret-are disclosed. We initiate the study of this new problem and introduce the notion of request complexity which captures the number of virtual network embedding requests needed to fully disclose the infrastructure topology. We derive lower bounds and present algorithms achieving an asymptotically optimal request complexity for the important class of tree and cactus graphs (complexity θ(n)) as well as arbitrary graphs (complexity θ(n2)).

Distributed recommender systems are becoming increasingly important for they address both scalability and the Big Brother syndrome. Link prediction is one of the core mechanism in recommender systems and relies on extracting some notion of proximity between entities in a graph. Applied to social networks, defining a proximity metric between users enable to predict potential relevant future relationships. In this paper, we propose SoCS (Social Coordinate Systems}, a fully distributed algorithm that embeds any social graph in an Euclidean space, which can easily be used to implement link prediction. To the best of our knowledge, SoCS is the first system explicitly relying on graph embedding. Inspired by recent works on non-isomorphic embeddings, the SoCS embedding preserves the community structure of the original graph, while being easy to decentralize. Nodes thus get assigned coordinates that reflect their social position. We show through experiments on real and synthetic data sets that these coordinates can be exploited for efficient link prediction.

It is well known that in an asynchronous system where processes are prone to crash, it is impossible to design a protocol that provides each process with the set of processes that are currently alive. Basically, this comes from the fact that it is impossible to distinguish a crashed process from a process that is very slow or with which communications are very slow. Nevertheless, designing protocols that provide the processes with good approximations of the set of processes that are currently alive remains a real challenge in fault-tolerant-distributed computing. This paper proposes such a protocol, plus a second protocol that allows to cope with heterogeneous communication networks. These protocols consider a realistic computation model where the processes are provided with nonsynchronized local clocks and a function \alpha () that takes a local duration \Delta as a parameter, and returns an integer that is an estimate of the number of processes that could have crashed during that duration \Delta. A simulation-based experimental evaluation of the proposed protocols is also presented. These experiments show that the protocols are practically relevant.

This note define self-structuring in a large-scale networked system as the ability of the participating entities to collaboratively impose a geometric structure to the network. This refers to assigning virtual coordinates to participating entities and to dividing the entities in several partitions, in such a way that each entity knows to which partition it belongs.

While electing an eventual common leader, despite process crashes, in a shared memory system where the processes communicate only by reading and writing shared registers is possible when the processes progress synchronously, this problem becomes impossible to solve as soon as the processes can progress in a fully asynchronous way. So, an important problem consists in finding additional behavioral assumptions that are, at the same time, "as weak as possible" (in order they are practically always satisfied), and "strong enough" in order to allow implementing an eventual leader service despite the net effect of asynchrony and failures. This paper focuses on this dilemma. More explicitly, it investigates a timing assumption that allows implementing an eventual leader in presence of partial asynchrony and process crashes. The proposed timing assumptions are particularly weak. They are the following: after some time (i) there is a process that behaves synchronously, and (ii) (t - f) other processes have timers that work correctly (t is the maximal number of processes that may crash, and f the actual number of process crashes; a timer works incorrectly when it expires too early with respect to the value it has been set). Then, the paper proposes a t-resilient protocol that elects an eventual common leader in any shared memory system that satisfies the previous assumption. Interestingly, this protocol is based on simple design principles.

No abstract available.

It is well-known that, in an asynchronous system where processes are prone to crash, it is impossible to design a protocol that provides each process with the set of processes that are currently alive. Basically, this comes from the fact that it is impossible to distinguish a crashed process from a process that is very slow or with which communications are very slow. Nevertheless, designing protocols that provide the processes with good approximations of the set of processes that are currently alive remains a real challenge in fault-tolerant distributed computing. This paper proposes such a protocol. To that end, it considers a realistic computation model where the processes are provided with non-synchronized local clocks and a function alpha(). That function takes a local duration as a parameter, and returns an integer that is an estimate of the number of processes that can crash during that duration. A simulation-based experimental evaluation of the protocol is also presented. The experiments show that the protocol is practically relevant