arXiv subjects

Gilles Tredan

Publications and source records attributed to Gilles Tredan.

At least 19 recordsLinked to original sources

arXiv2026

Token-Efficient Change Detection in LLM APIs

Remote change detection in LLMs is a difficult problem. Existing methods are either too expensive for deployment at scale, or require initial white-box access to model weights or grey-box access to log probabilities. We aim to achieve both low cost and strict black-box operation, observing only output tokens. Our approach hinges on specific inputs we call Border Inputs, for which there exists more than one output top token. From a statistical perspective, optimal change detection depends on the model's Jacobian and the Fisher information of the output distribution. Analyzing these quantities in low-temperature regimes shows that border inputs enable powerful change detection tests. Building on this insight, we propose the Black-Box Border Input Tracking (B3IT) scheme. Extensive in-vivo and in-vitro experiments show that border inputs are easily found for non-reasoning tested endpoints, and achieve performance on par with the best available grey-box approaches. B3IT reduces costs by $30\times$ compared to existing methods, while operating in a strict black-box setting.

cs.LG↗

arXiv2026

FLIPS: Instance-Fingerprinting for LLMs via Pseudo-random Sequences

Literature reveals that a Large Language Model's (LLM) behavior is not only conditioned by its original weights but also its instance-level parameters, such as instructional prompt, sampling configuration or quantization. A model that generates safe outputs under one configuration may produce toxic content under another. However, current LLM identification techniques (such as fingerprinting) focus on intellectual property protection, and their design favors robustness to changes in these instance-level parameters. This poses a critical challenge for AI regulation in which compliance assessments target actual deployed behaviors, not model provenance. In this paper, we introduce instance-level fingerprinting, a regulator-oriented paradigm that distinguishes configurations of the same LLM. Our method FLIPS, exploits biases in generated binary random sequences to reach 96% (closed-set) and 90% (open-set, where some targets are unknown) identification accuracy across 237 model instances, versus 35% for the adapted LLMmap baseline. This shows that instance-level fingerprinting is both necessary for regulation and practically feasible. Code available at https://github.com/GurvanR/FLIPS-LLM-Instance-Fingerprinting.

cs.LG↗

arXiv2026

Leveraging Imperfect Sources to Detect Fairwashing in Black-Box Auditing

Algorithmic auditing has become central to platform accountability under frameworks such as the AI Act and the Digital Services Act. In practice, this obligation is discharged through dedicated Audit APIs. This architecture creates a paradox: the entity under scrutiny controls the evaluation interface. A platform facing legal sanctions can serve a compliant surrogate model on its Audit API, while running a discriminatory production system. This deceptive practice is known as fairwashing. Manipulation is undetectable if the auditor relies on only one source. To address this limitation, we introduce the Two-Source Audit Model (2SAM). This model cross-references the Audit API with an independent trusted stream. The key insight is that the trusted stream does not need to be perfectly aligned with the Audit API. We introduce a consistency proxy, a probabilistic mapping that can reconcile discrepancies between sources. This approach yields three results. First, we quantify the rate of manipulation above which a single-source auditor is blind. Second, we show how proxy quality governs detection power. Third, we provide a closed-form budget condition guaranteeing detection at any target confidence level, closing the blind spot mentioned above. We validate 2SAM on the UCI Adult dataset, achieving $70\%$ detection power with as few as $127$ cross-verification queries out of a total budget of $750$, using a name-based gender proxy with $94.2\%$ accuracy.

cs.LG↗

arXiv2026

Log Probability Tracking of LLM APIs

When using an LLM through an API provider, users expect the served model to remain consistent over time, a property crucial for the reliability of downstream applications and the reproducibility of research. Existing audit methods are too costly to apply at regular time intervals to the wide range of available LLM APIs. This means that model updates are left largely unmonitored in practice. In this work, we show that while LLM log probabilities (logprobs) are usually non-deterministic, they can still be used as the basis for cost-effective continuous monitoring of LLM APIs. We apply a simple statistical test based on the average value of each token logprob, requesting only a single token of output. This is enough to detect changes as small as one step of fine-tuning, making this approach more sensitive than existing methods while being 1,000x cheaper. We introduce the TinyChange benchmark as a way to measure the sensitivity of audit methods in the context of small, realistic model changes.

cs.LG↗

arXiv2025

Robust ML Auditing using Prior Knowledge

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.

cs.LG↗

arXiv2025

LLMs Prompted for Graphs: Hallucinations and Generative Capabilities

Large Language Models (LLMs) are nowadays prompted for a wide variety of tasks. In this article, we investigate their ability in reciting and generating graphs. We first study the ability of LLMs to regurgitate well known graphs from the literature (e.g. Karate club or the graph atlas)4. Secondly, we question the generative capabilities of LLMs by asking for Erdos-Renyi random graphs. As opposed to the possibility that they could memorize some Erdos-Renyi graphs included in their scraped training set, this second investigation aims at studying a possible emergent property of LLMs. For both tasks, we propose a metric to assess their errors with the lens of hallucination (i.e. incorrect information returned as facts). We most notably find that the amplitude of graph hallucinations can characterize the superiority of some LLMs. Indeed, for the recitation task, we observe that graph hallucinations correlate with the Hallucination Leaderboard, a hallucination rank that leverages 10, 000 times more prompts to obtain its ranking. For the generation task, we find surprisingly good and reproducible results in most of LLMs. We believe this to constitute a starting point for more in-depth studies of this emergent capability and a challenging benchmark for their improvements. Altogether, these two aspects of LLMs capabilities bridge a gap between the network science and machine learning communities.

cs.CL↗

arXiv2025

P2NIA: Privacy-Preserving Non-Iterative Auditing

The emergence of AI legislation has increased the need to assess the ethical compliance of high-risk AI systems. Traditional auditing methods rely on platforms' application programming interfaces (APIs), where responses to queries are examined through the lens of fairness requirements. However, such approaches put a significant burden on platforms, as they are forced to maintain APIs while ensuring privacy, facing the possibility of data leaks. This lack of proper collaboration between the two parties, in turn, causes a significant challenge to the auditor, who is subject to estimation bias as they are unaware of the data distribution of the platform. To address these two issues, we present P2NIA, a novel auditing scheme that proposes a mutually beneficial collaboration for both the auditor and the platform. Extensive experiments demonstrate P2NIA's effectiveness in addressing both issues. In summary, our work introduces a privacy-preserving and non-iterative audit scheme that enhances fairness assessments using synthetic or local data, avoiding the challenges associated with traditional API-based audits.

cs.LG↗

arXiv2024

The 20 questions game to distinguish large language models

In a parallel with the 20 questions game, we present a method to determine whether two large language models (LLMs), placed in a black-box context, are the same or not. The goal is to use a small set of (benign) binary questions, typically under 20. We formalize the problem and first establish a baseline using a random selection of questions from known benchmark datasets, achieving an accuracy of nearly 100% within 20 questions. After showing optimal bounds for this problem, we introduce two effective questioning heuristics able to discriminate 22 LLMs by using half as many questions for the same task. These methods offer significant advantages in terms of stealth and are thus of interest to auditors or copyright owners facing suspicions of model leaks.

cs.CL↗

arXiv2024

Fairness Auditing with Multi-Agent Collaboration

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.

cs.LG↗

arXiv2024

Under manipulations, are some AI models harder to audit?

Auditors need robust methods to assess the compliance of web platforms with the law. However, since they hardly ever have access to the algorithm, implementation, or training data used by a platform, the problem is harder than a simple metric estimation. Within the recent framework of manipulation-proof auditing, we study in this paper the feasibility of robust audits in realistic settings, in which models exhibit large capacities. We first prove a constraining result: if a web platform uses models that may fit any data, no audit strategy -- whether active or not -- can outperform random sampling when estimating properties such as demographic parity. To better understand the conditions under which state-of-the-art auditing techniques may remain competitive, we then relate the manipulability of audits to the capacity of the targeted models, using the Rademacher complexity. We empirically validate these results on popular models of increasing capacities, thus confirming experimentally that large-capacity models, which are commonly used in practice, are particularly hard to audit robustly. These results refine the limits of the auditing problem, and open up enticing questions on the connection between model capacity and the ability of platforms to manipulate audit attempts.

cs.LG↗

arXiv2023

Modeling Rabbit-Holes on YouTube

Numerous discussions have advocated the presence of a so called rabbit-hole (RH) phenomenon on social media, interested in advanced personalization to their users. This phenomenon is loosely understood as a collapse of mainstream recommendations, in favor of ultra personalized ones that lock users into narrow and specialized feeds. Yet quantitative studies are often ignoring personalization, are of limited scale, and rely on manual tagging to track this collapse. This precludes a precise understanding of the phenomenon based on reproducible observations, and thus the continuous audits of platforms. In this paper, we first tackle the scale issue by proposing a user-sided bot-centric approach that enables large scale data collection, through autoplay walks on recommendations. We then propose a simple theory that explains the appearance of these RHs. While this theory is a simplifying viewpoint on a complex and planet-wide phenomenon, it carries multiple advantages: it can be analytically modeled, and provides a general yet rigorous definition of RHs. We define them as an interplay between i) user interaction with personalization and ii) the attraction strength of certain video categories, which cause users to quickly step apart of mainstream recommendations made to fresh user profiles. We illustrate these concepts by highlighting some RHs found after collecting more than 16 million personalized recommendations on YouTube. A final validation step compares our automatically-identified RHs against manually-identified RHs from a previous research work. Together, those results pave the way for large scale and automated audits of the RH effect in recommendation systems.

cs.CY↗

arXiv2022

On the Price of Locality in Static Fast Rerouting

Modern communication networks feature fully decentralized 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. (ACM PODC'12) and Foerster et al. (SIAM APOCS'21) 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.

cs.DC↗

arXiv2020

On the Feasibility of Perfect Resilience with Local Fast Failover

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. These rules determine for each incoming link from which a packet may arrive and the set of local link failures (i.e., the failed links incident to a node), on which outgoing link a packet should be forwarded. 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. (ACM PODC 2012) and also Chiesa et al. (IEEE/ACM Trans. Netw. 2017) showed that it is not always possible to provide perfect resilience. Interestingly, not much more is known currently about the feasibility of perfect resilience. This paper revisits perfect resilience with local fast failover, both in a model where the source can and cannot be used for forwarding decisions. We first derive several fairly general impossibility results: By establishing a connection between graph minors and resilience, we prove that it is impossible to achieve perfect resilience on any non-planar graph; furthermore, while planarity is necessary, it is also not sufficient for perfect resilience. On the positive side, we show that graph families closed under link subdivision allow for simple and efficient failover algorithms which simply skip failed links. We demonstrate this technique by deriving perfect resilience for outerplanar graphs and related scenarios, as well as for scenarios where the source and target are topologically close after failures.

cs.NI↗

arXiv2020

The Bouncer Problem: Challenges to Remote Explainability

The concept of explainability is envisioned to satisfy society's demands for transparency on machine learning decisions. The concept is simple: like humans, algorithms should explain the rationale behind their decisions so that their fairness can be assessed. While this approach is promising in a local context (e.g. to explain a model during debugging at training time), we argue that this reasoning cannot simply be transposed in a remote context, where a trained model by a service provider is only accessible through its API. This is problematic as it constitutes precisely the target use-case requiring transparency from a societal perspective. Through an analogy with a club bouncer (which may provide untruthful explanations upon customer reject), we show that providing explanations cannot prevent a remote service from lying about the true reasons leading to its decisions. More precisely, we prove the impossibility of remote explainability for single explanations, by constructing an attack on explanations that hides discriminatory features to the querying user. We provide an example implementation of this attack. We then show that the probability that an observer spots the attack, using several explanations for attempting to find incoherences, is low in practical settings. This undermines the very concept of remote explainability in general.

cs.LG↗

arXiv2019

Collective Information Processing in Human Phase Separation

Social media filters combined with recommender systems can lead to the emergence of filter bubbles and polarized groups. In addition, segregation processes of human groups in certain social contexts have been shown to share some similarities with phase separation phenomena in physics. Here, we study the impact of information filtering on collective segregation behavior. We report a series of experiments where groups of 22 subjects have to perform a collective segregation task that mimics the tendency of individuals to bond with other similar individuals. More precisely, the participants are each assigned a color (red or blue) unknown to them, and have to regroup with other subjects sharing the same color. To assist them, they are equipped with an artificial sensory device capable of detecting the majority color in their ``environment'' (defined as their $k$ nearest neighbors, unbeknownst to them), for which we control the perception range, $k=1,3,5,7,9,11,13$. We study the separation dynamics (emergence of unicolor groups) and the properties of the final state, and show that the value of $k$ controls the quality of the segregation, although the subjects are totally unaware of the precise definition of the ``environment''. We also find that there is a perception range $k=7$ above which the ability of the group to segregate does not improve. We introduce a model that precisely describes the random motion of a group of pedestrians in a confined space, and which faithfully reproduces and allows to interpret the results of the segregation experiments. Finally, we discuss the strong and precise analogy between our experiment and the phase separation of two immiscible materials at very low temperature.

physics.soc-ph↗

arXiv2019

TamperNN: Efficient Tampering Detection of Deployed Neural Nets

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.

cs.CR↗

arXiv2017

The Many Faces of Graph Dynamics

The topological structure of complex networks has fascinated researchers for several decades, resulting in the discovery of many universal properties and reoccurring characteristics of different kinds of networks. However, much less is known today about the network dynamics: indeed, complex networks in reality are not static, but rather dynamically evolve over time. Our paper is motivated by the empirical observation that network evolution patterns seem far from random, but exhibit structure. Moreover, the specific patterns appear to depend on the network type, contradicting the existence of a "one fits it all" model. However, we still lack observables to quantify these intuitions, as well as metrics to compare graph evolutions. Such observables and metrics are needed for extrapolating or predicting evolutions, as well as for interpolating graph evolutions. To explore the many faces of graph dynamics and to quantify temporal changes, this paper suggests to build upon the concept of centrality, a measure of node importance in a network. In particular, we introduce the notion of centrality distance, a natural similarity measure for two graphs which depends on a given centrality, characterizing the graph type. Intuitively, centrality distances reflect the extent to which (non-anonymous) node roles are different or, in case of dynamic graphs, have changed over time, between two graphs. We evaluate the centrality distance approach for five evolutionary models and seven real-world social and physical networks. Our results empirically show the usefulness of centrality distances for characterizing graph dynamics compared to a null-model of random evolution, and highlight the differences between the considered scenarios. Interestingly, our approach allows us to compare the dynamics of very different networks, in terms of scale and evolution speed.

cs.SI↗

arXiv2016

The Many Faces of Graph Dynamics

cs.SI↗