This paper addresses detection of a reverse engineering (RE) attack targeting a deep neural network (DNN) image classifier; by querying, RE's aim is to discover the classifier's decision rule. RE can enable test-time evasion attacks, which require knowledge of the classifier. Recently, we proposed a quite effective app…
On-device research index
arXiv research
A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.
168,982 papers · 148 categories
Trend · papers per month
48 results for “test-time evasion”
New model improves classifier security against evasion attacks by selecting features.
problem Security of machine learning classifiers against evasion attacks.
method Adversary-aware feature selection model incorporating specific adversary assumptions.
result Classifier security worsens with feature selection, but new model improves it.
Why Do Adversarial Attacks Transfer? Explaining Transferability of Evasion and Poisoning Attackscs.LG
Transferability captures the ability of an attack against a machine-learning model to be effective against a different, potentially unknown, model. Empirical evidence for transferability has been shown in previous work, but the underlying reasons why an attack transfers or not are not yet well understood. In this paper…
This paper reviews defenses against adversarial learning attacks on statistical classifiers.
problem Adversarial attacks on machine learning systems, particularly statistical classifiers.
method Survey of recent work on test-time evasion, data poisoning, and reverse engineering attacks and defenses.
result Novel insights that challenge conventional AL wisdom and target unresolved issues.
Deep neural networks (DNNs) have transformed several artificial intelligence research areas including computer vision, speech recognition, and natural language processing. However, recent studies demonstrated that DNNs are vulnerable to adversarial manipulations at testing time. Specifically, suppose we have a testing …
New approach verifies tree ensembles efficiently, making them more robust.
problem Verifying robustness of decision tree ensembles against evasion attacks is NP-hard.
method Identified large-spread ensembles, proposed verifiable learning, designed training algorithm.
result Large-spread ensembles can be verified in polynomial time and are more robust.
Feature Partitioning for Robust Tree Ensembles and their Certification in Adversarial Scenarioscs.LG
This work proposes a robust ensemble method for decision trees that resists adversarial attacks.
problem Adversarial attacks on machine learning models, especially decision trees.
method Feature partitioning to train robust ensembles and approximate certification methods.
result The proposed ensemble method can resist evasion attacks by a majority of its models.
2DSig-Detect: a semi-supervised framework for anomaly detection on image data using 2D-signaturescs.CV
2DSig-Detect detects adversarial perturbations in images.
problem Adversarial attacks degrade image model performance.
method 2D-signature embedded semi-supervised framework using rough path theory.
result 2DSig-Detect outperforms other methods in detecting adversarial perturbations.
In machine learning (ML) security, attacks like evasion, model stealing or membership inference are generally studied in individually. Previous work has also shown a relationship between some attacks and decision function curvature of the targeted model. Consequently, we study an ML model allowing direct control over t…
Reviews spams are prevalent in e-commerce to manipulate product ranking and customers decisions maliciously. While spams generated based on simple spamming strategy can be detected effectively, hardened spammers can evade regular detectors via more advanced spamming strategies. Previous work gave more attention to evas…
Making learners robust to adversarial perturbation at test time (i.e., evasion attacks) or training time (i.e., poisoning attacks) has emerged as a challenging task. It is known that for some natural settings, sublinear perturbations in the training phase or the testing phase can drastically decrease the quality of the…
Microscopic models describing a whole of economic interactions in a closed society are considered. The presence of a tax system combined with a redistribution process is taken into account, as well as the occurrence of tax evasion. In particular, the existence is postulated, in relation to the level of evasion, of diff…
Stochastic compliance/evasion dynamics in tax models: a piecewise deterministic Markov process approachphysics.soc-ph
New model predicts dynamic tax evasion with audits and imitation.
problem Static treatment of tax compliance and evasion in Bertotti and Modanese model.
method Piecewise Deterministic Markov Processes (PDMPs) for audits and imitation mechanisms.
result Model shows persistent fluctuations and stationary distribution, not extreme equilibrium.
We develop a model of tax evasion based on the Ising model. We augment the model using an appropriate enforcement mechanism that may allow policy makers to curb tax evasion. With a certain probability tax evaders are subject to an audit. If they get caught they behave honestly for a certain number of periods. Simulatin…
This paper explores evasion attacks against Bayesian models.
problem Bayesian predictive models are vulnerable to evasion attacks.
method Developed gradient-based attacks for specific point predictions and entire posterior distributions.
result Optimal evasion attacks can be designed against Bayesian models.
Treant creates evasion-resistant decision trees.
problem Evasion attacks on decision tree ensembles.
method Treant minimizes evasion-aware loss function, using robust splitting and attack invariance.
result Treant generates accurate, nearly evasion-insensitive decision trees.
We analyze income tax evasion dynamics in a standard model of statistical mechanics, the Ising model of ferromagnetism. However, in contrast to previous research, we use an inhomogeneous multi-dimensional Ising model where the local degrees of freedom (agents) are subject to a specific social temperature and coupled to…
Study malware evasion attacks and defenses using ML classifiers.
problem Adversarial examples can mislead ML-based malware detectors.
method Investigated white-box and grey-box evasion attacks; compared defense approaches.
result Proposed a framework for grey-box and black-box attacks.
Study of pursuit-evasion game on sphere and its relation to planar Apollonius circle.
problem Analyzing pursuit-evasion game on a sphere and its properties.
method Extending classical planar pursuit-evasion game to spherical geometry, studying equilibrium intercept points and their relation to Apollonius domain.
result Condition for intercept point to belong to Apollonius domain on sphere, analogous to planar game.
Micro to macro models for income distribution in the absence and in the presence of tax evasionq-fin.GN
We investigate the effect of tax evasion on the income distribution and the inequality index of a society through a kinetic model described by a set of nonlinear ordinary differential equations. The model allows to compute the global outcome of binary and multiple microscopic interactions between individuals. When evas…
We investigate an inhomogeneous Ising model in the context of tax evasion dynamics where different types of agents are parametrized via local temperatures and magnetic fields. In particular, we analyse the impact of backauditing and endogenously determined penalty rates on tax compliance. Both features contribute to a …
Pursuit-evasion is a multi-agent sequential decision problem wherein a group of agents known as pursuers coordinate their traversal of a spatial domain to locate an agent trying to evade them. Pursuit evasion problems arise in a number of import application domains including defense and route planning. Learning to opti…
Computes sections of a submersion and applies to evasion path problem.
problem Evasion path problem for mobile sensor networks.
method Computation of sections from fiber homotopy groups and time-varying homology/cohomology.
result Necessary and sufficient conditions for evasion paths and lower bounds.
The paper certifies decision trees against evasion attacks using program analysis.
problem Vulnerability of decision tree models to evasion attacks by maliciously crafted perturbations.
method Transform decision trees into imperative programs for program analysis, leveraging abstract interpretation.
result Soundly verifies security guarantees of decision tree models, yielding minimal false positives.
Adversarial attacks can fool self-driving cars, changing steering predictions.
problem Security of deep neural networks in self-driving cars.
method Demonstrated adversarial attacks on steering angle prediction model.
result Small image modifications can mislead self-driving car predictions.
Tool detects tax evasion on social media using multi-modal deep learning.
problem Detecting tax evasion on social media platforms.
method Developed a multi-modal deep neural network combining comments, hashtags, and images.
result Multi-modal deep neural network achieved AUC of 0.808 and F1 score of 0.762.
Classifier evasion consists in finding for a given instance the nearest instance such that the classifier predictions of and are different. We present two novel algorithms for systematically computing evasions for tree ensembles such as boosted trees and random forests. Our first algorithm uses a Mixe…
The existence of evasion attacks during the test phase of machine learning algorithms represents a significant challenge to both their deployment and understanding. These attacks can be carried out by adding imperceptible perturbations to inputs to generate adversarial examples and finding effective defenses and detect…
We consider the static and dynamic models of Cournot duopoly with tax evasion. In the dynamic model we introduce the time delay and we analyze the local stability of the stationary state. There is a critical value of the delay when the Hopf bifurcation occurs.
Open-world evasion attacks using OOD adversarial examples outperform in-distribution attacks.
problem Evasion attacks on deep learning systems in open-world environments.
method Using 11 neural network models trained on 3 datasets, the study demonstrates the effectiveness of OOD adversarial examples against state-of-the-art detectors and defenses.
result OOD adversarial examples can achieve up to 4x higher success rates than in-distribution adversarial examples against known defenses.
PAD offers a principled approach to malware detection against evasion attacks.
problem Machine Learning techniques for malware detection are vulnerable to evasion attacks.
method PAD proposes a new adversarial training framework with convergence guarantees for robust optimization.
result PAD significantly outperforms state-of-the-art defenses and can harden ML-based malware detection against 27 evasion attacks.
secml is a Python library for secure and explainable machine learning.
problem Evaluate and understand security and explainability of machine learning models.
method Implement attacks and explainability methods for various algorithms.
result Demonstrates how quickly models fail under adversarial perturbations and visualizes influential features.
The convolutional neural network (CNN) architecture is increasingly being applied to new domains, such as malware detection, where it is able to learn malicious behavior from raw bytes extracted from executables. These architectures reach impressive performance with no feature engineering effort involved, but their rob…
Efficient black-box evasion attacks against deep networks with fewer queries.
problem Optimizing targeted evasion attacks with limited query budget.
method Formalized problem, evaluated benefits of using substitute models, presented two new attack strategies.
result Attack strategies are as effective as previous techniques but require significantly fewer queries.
Generative ML learns optimal pursuit trajectories in pursuit-evasion games.
problem Optimizing Blue's pursuit trajectory to intercept Red in a game of pursuit-evasion.
method Applying generative machine learning to learn optimal action policies for Blue.
result Generative ML models can learn relevant representations for pursuit-evasion dynamics.
In 1967, Chillingworth proved that all convex simplicial 3-balls are collapsible. Using the classical notion of tightness, we generalize this to arbitrary manifolds: We show that all tight simplicial 3-manifolds admit some perfect discrete Morse function. We also strengthen Chillingworth's theorem by proving that all c…
This paper improves neural network robustness through feedback learning.
problem Vulnerability of neural networks to adversarial attacks.
method Feedback learning method to understand and improve model robustness.
result Significantly improved models' accuracy and robustness against evasion attacks.
This work studies PAC learning under evasion attacks, proving exponential sample complexity for high-dimensional inputs.
problem Formal study of PAC learning under evasion attacks where the adversary misclassifies perturbed samples.
method Proves exponential sample complexity for high-dimensional inputs under evasion attacks, formalizes hybrid attacks.
result PAC learning requires exponential sample complexity for high-dimensional inputs under evasion attacks.
UAVs use RL to outperform a greedy shooter in 2D engagements.
problem Optimizing UAV pursuit-evasion strategies in 2D environments.
method Reinforcement learning applied to a greedy shooter strategy in a 2D aerial engagement.
result Deep Q-learning UAV outperforms greedy shooter UAV in simulations.
We calculate the dynamics of tax evasion within a multi-agent econophysics model which is adopted from the theory of magnetism and previously has been shown to capture the main characteristics from agent-based based models which build on the standard Allingham and Sandmo approach. In particular, we implement a feedback…
The paper calculates how much data is needed to learn decision lists in the presence of evasion attacks.
problem Quantifying sample complexity for robust learning of decision lists against evasion attacks.
method PAC learning framework, Lipschitz condition on distributions, lower and upper bounds on sample complexity.
result Upper and lower bounds on sample complexity for robust learning of decision lists, showing exponential vs polynomial dependence on adversary's budget.
Paper uses 2-step Gradient Boosting to predict VAT tax gap.
problem Estimating tax evasion and revenue loss from tax avoidance.
method 2-steps Gradient Boosting model to correct selection bias.
result Significantly improved prediction of VAT tax gap.
Study finds tax avoidance and IT issues hinder revenue in Gombe state.
problem Problems of personal income tax on revenue generation in Gombe state.
method Survey with primary and secondary data, chi square test.
result Tax avoidance and IT issues are major problems.
Traditional classification algorithms assume that training and test data come from similar distributions. This assumption is violated in adversarial settings, where malicious actors modify instances to evade detection. A number of custom methods have been developed for both adversarial evasion attacks and robust learni…
Deep Convolutional Networks (DCNs) have been shown to be vulnerable to adversarial examples---perturbed inputs specifically designed to produce intentional errors in the learning algorithms at test time. Existing input-agnostic adversarial perturbations exhibit interesting visual patterns that are currently unexplained…
This study evaluates the robustness of transformation-based ensemble defense against evasion attacks.
problem Understanding the reasons behind the robustness improvement in transformation-based ensemble defense.
method Designing two adaptive attacks to evaluate transformation-based ensemble defense, conducting experiments to analyze robustness.
result The robustness improvement is mainly from irreversible transformations rather than the ensemble of models.
We construct the first explicit example of a simplicial 3-ball B_{15,66} that is not collapsible. It has only 15 vertices. We exhibit a second 3-ball B_{12,38} with 12 vertices that is collapsible and evasive, but not shellable. Finally, we present the first explicit triangulation of a 3-sphere S_{18, 125} (with only 1…
Paper defends models from backdoor attacks using randomized smoothing.
problem Vulnerability of deep neural networks to backdoor attacks.
method Unified framework via randomized smoothing for both evasion and backdoor attacks; RAB robust training process.
result First provable robustness bound for backdoor attacks.