Adversarial Learning Techniques Research Articles

Clinical decision making under uncertainty: a bootstrapped counterfactual inference approach

BackgroundLearning policies for decision-making, such as recommending treatments in clinical settings, is important for enhancing clinical decision-support systems. However, the challenge lies in accurately evaluating and optimizing these policies for maximum efficacy. This paper addresses this gap by focusing on two key aspects of policy learning: evaluation and optimization.MethodWe develop counterfactual policy learning algorithms for practical clinical applications to suggest viable treatment for patients. We first design a bootstrap method for counterfactual assessment and enhancement of policies, aiming to diminish uncertainty in clinical decisions. Building on this, we introduce an innovative adversarial learning algorithm, inspired by bootstrap principles, to further advance policy optimization.ResultsThe efficacy of our algorithms was validated using both semi-synthetic and real-world clinical datasets. Our method outperforms baseline algorithms, reducing the variance in policy evaluation by 30% and the error rate by 25%. In policy optimization, it enhances the reward by 1% to 3%, highlighting the practical value of our approach in clinical decision-making.ConclusionThis study demonstrates the effectiveness of combining bootstrap and adversarial learning techniques in policy learning for clinical decision support. It not only enhances the accuracy and reliability of policy evaluation and optimization but also paves avenues for leveraging advanced counterfactual machine learning in healthcare.

Unsupervised Transfer Learning Across Different Data Modalities for Bearing's Speed Identification

Recent advancements in transfer learning have revolutionized predictive maintenance, enabling cross-domain generalization for components with varying characteristics and operating under different conditions. While traditional transfer learning approaches require labeled data in both source and target domains, unsupervised transfer learning strives for a more cost-effective alternative for which only labels are available in the source domain. This study investigates adversarial transfer learning between two different sensor modalities: vibration and acoustic. The goal is to enable bearing monitoring using microphones, which are, in general terms, cheaper and easier to deploy than vibration sensors; and without the need to label data in the target domain. The research goal is to identify the operating speed of a bearing testbed. The source domain data correspond to vibration measurements taken from an attached sensor, while the target domain uses a microphone array at distance. Artificial Neural Networks are used as the base architecture. Transferability is assessed with two unsupervised adversarial learning techniques: gradient reversal and deep correlation alignment. Their performance is compared to traditional supervised transfer learning via fine-tuning. Experimental results demonstrate that gradient reversal outperforms deep correlation alignment and is able to achieve results similar to those obtained with supervised transfer learning. These findings highlight the feasibility of speed identification using a microphone array and establish a baseline for future condition monitoring research with such sensors. Conclusion. These findings highlight the feasibility of speed identification using a microphone array and establish a baseline for future condition monitoring research with such sensors.

Adversarial Learning for Constrained Image Splicing Detection and Localization

Abstract: In the era of digital media, the manipulation of images has become a significant concern, particularly on social media platforms. Adversarial Learning for Constrained Image Splicing Detection and Localization aims to develop a robust system capable of recognizing and localizing spliced or tampered images. This research proposes a novel approach that leverages Convolutional Neural Networks (CNNs) and adversarial learningtechniques to enhance the detection and localization of image splicing. The system is designed to aid in distinguishing between authentic and manipulated images, thereby promoting transparency and trust in digital media. With the rapid advancement of digital media and image editingtools, the manipulation of visual content has become increasingly prevalent, posing significant challenges to the credibility and trustworthiness of images shared on social media platforms. One pervasive form of image manipulation is image splicing, where portions of two or more images are seamlessly combined to create a composite image, often with the intent to mislead or deceive viewers. Adversarial Learning for Constrained Image Splicing Detection and Localization aims to address this issue by developing a robust system capable of accurately detecting and localizing spliced regions within tampered images. This research proposes a novel approach that synergistically integrates Convolutional Neural Networks (CNNs) and adversarial learning techniques to enhance the performance of image splicing detection and localization. The proposed system employs a two-stage process: first, a CNN-based classifier is trained on a large dataset of authentic and spliced images to learn discriminative features for distinguishing between the two classes; subsequently, an adversarial learning model is employed to generate adversarial examples that can deceive the CNN classifier, while the classifier is iteratively updated to become more robust against these adversarial perturbations. By accurately identifying and localizing spliced regions within images, this research contributes to promoting transparency and trust in digital media shared on social media platforms, ultimately empowering users to make informed decisions about the authenticity of visual content. The proposed system has potential applications in various domains, including journalism, law enforcement, and content moderation, where verifying the integrity of visual evidence is of paramount importance.

Fairness without Demographics through Shared Latent Space-Based Debiasing

Ensuring fairness in machine learning (ML) is crucial, particularly in applications that impact diverse populations. The majority of existing works heavily rely on the availability of protected features like race and gender. However, practical challenges such as privacy concerns and regulatory restrictions often prohibit the use of this data, limiting the scope of traditional fairness research. To address this, we introduce a Shared Latent Space-based Debiasing (SLSD) method that transforms data from both the target domain, which lacks protected features, and a separate source domain, which contains these features, into correlated latent representations. This allows for joint training of a cross-domain protected group estimator on the representations. We then debias the downstream ML model with an adversarial learning technique that leverages the group estimator. We also present a relaxed variant of SLSD, the R-SLSD, that occasionally accesses a small subset of protected features from the target domain during its training phase. Our extensive experiments on benchmark datasets demonstrate that our methods consistently outperform existing state-of-the-art models in standard group fairness metrics.

Deep Learning–based Dynamic User Alignment in Social Networks

Academics and businesses are paying intense attention to social network alignment, which centers various social networks around their shared members. All studies to date treat the social network as static and ignore its innate dynamism. In reality, an individual's discriminative pattern is embedded in the dynamics of social networks, and this information may be used to improve social network alignment. This study finds that these dynamics can reveal more apparent patterns better suited to lining up the social web of things (SWoT). The correlation between the user structure and attributes for each social network must be maintained to combine the binary dynamics and make the original synthetic embedding representation. Finally, the initial embedding of each network is projected to a target subspace as part of the semi-supervised spatial transformation learning process. The Dynamic Social Network Alignment approach outperforms the current mainstream algorithm by 10% in this article's extensive series of trials using real-world datasets. The findings of this study show that this alignment of enormous networks addresses the volume, variety, velocity, and veracity (or 4Vs) of vast networks. To improve the efficacy and resilience of an adversarial network alignment, adversarial learning techniques can be applied. The results show that the model with structure, attribute, and time information performs the best, while the model without attribute information comes in second, the model without time information performs mediocrely, and the model without structure information performs the worst.

PURE: Passive Multi-Person Identification via Footstep for Mobile Service Networks

Recently, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">passive behavioral biometrics</i> (e.g., gesture or footstep) acquired from wireless networks or mobile services have become promising complements to conventional user identification methods (e.g., face or fingerprint) under special situations, yet existing sensing technologies require lengthy measurement traces and cannot identify multiple users at the same time. To this end, we propose PURE as a passive multi-person identification system leveraging deep learning enabled footstep separation and recognition. PURE passively identifies a user by identifying the unique “footprints” in its footstep. Different from existing gait-enabled recognition systems incurring a long sensing delay to acquire many footsteps, PURE can recognize a person by as few as only one step, substantially cutting the identification latency. To make PURE adaptive to walking pace variations, environmental dynamics, and even unseen targets, we apply an adversarial learning technique to improve its domain generalisability and identification accuracy. Finally, PURE is robust against replay attack, enabled by the richness of footstep and spatial awareness. We implement a PURE prototype using commodity hardware and evaluate it in typical indoor settings. Evaluation results demonstrate a cross-domain identification accuracy of over 90%.

Multi-storey shear type buildings under earthquake loading: Adversarial learning-based prediction of the transient dynamics and damage classification

In this paper, the transient dynamic response of shear type multi-storey buildings subjected to earthquake ground motion is generated via adversarial learning technique under different damage conditions, starting from the relevant undamaged responses. A Representation Generative Adversarial Network (RepGAN) is trained on a database of synthetic accelerograms to obtain the responses of the buildings in their undamaged state and in case of plausible damage patterns. Each structural response, represented by a set of time histories to catch the lateral storey displacements/accelerations, is encoded to learn its hidden features and infer the associated damage class. By re-sampling the encoded latent space, it is shown how to switch from the undamaged to the damaged class and to decode the damaged response. The proposed methodology enables damage classification in shear-type multi-storey buildings proving that it can successfully detect damage and assess two different damage severity levels whenever the time-history of a sufficient number of floors is monitored. To outline the generalization capability of the proposed approach, the signal reconstruction is quantitatively assessed for all damage conditions and even in case of a damage condition different from the one corresponding to the encoded signal.

Minimum Adversarial Distribution Discrepancy for Domain Adaptation

Domain adaptation (DA) refers to generalize a learning technique across the source domain and target domain under different distributions. Therefore, the essential problem in DA is how to reduce the distribution discrepancy between the source and target domains. Typical methods are to embed the adversarial learning technique into deep networks to learn transferable feature representations. However, existing adversarial related DA methods may not sufficiently minimize the distribution discrepancy. In this article, a DA method minimum adversarial distribution discrepancy (MADD) is proposed by combining feature distribution with adversarial learning. Specifically, we design a novel divergence metric loss, named maximum mean discrepancy based on conditional entropy (MMD-CE), and embed it in the adversarial DA network. The proposed MMD-CE loss can address two problems: 1) the misalignment from different class distributions between domains and 2) the equilibrium challenge issue in adversarial DA. Comparative experiments on Office-31, ImageCLEF-DA, and Office-Home data sets with state-of-the-art methods show that our method has some advantageous performances.

BV-GAN: 3D time-of-flight magnetic resonance angiography cerebrovascular vessel segmentation using adversarial CNNs.

Purpose: Cerebrovascular vessel segmentation is a key step in the detection of vessel pathology. Brain time-of-flight magnetic resonance angiography (TOF-MRA) is a main method used clinically for imaging of blood vessels using magnetic resonance imaging. This method is primarily used to detect narrowing, blockage of the arteries, and aneurysms. Despite its importance, TOF-MRA interpretation relies mostly on visual, subjective assessment performed by a neuroradiologist and is mostly based on maximum intensity projections reconstruction of the three-dimensional (3D) scan, thus reducing the acquired spatial resolution. Works tackling the central problem of automatically segmenting brain blood vessels typically suffer from memory and imbalance related issues. To address these issues, the spatial context of the segmentation consider by neural networks is typically restricted (e.g., by resolution reduction or analysis of environments of lower dimensions). Although efficient, such solutions hinder the ability of the neural networks to understand the complex 3D structures typical of the cerebrovascular system and to leverage this understanding for decision making. Approach: We propose a brain-vessels generative-adversarial-network (BV-GAN) segmentation model, that better considers connectivity and structural integrity, using prior based attention and adversarial learning techniques. Results: For evaluations, fivefold cross-validation experiments were performed on two datasets. BV-GAN demonstrates consistent improvement of up to 10% in vessel Dice score with each additive designed component to the baseline state-of-the-art models. Conclusions: Potentially, this automated 3D-approach could shorten analysis time, allow for quantitative characterization of vascular structures, and reduce the need to decrease resolution, overall improving diagnosis cerebrovascular vessel disorders.

Effect of Kinematics and Fluency in Adversarial Synthetic Data Generation for ASL Recognition With RF Sensors

RF sensors have been recently proposed as a new modality for sign language processing technology. They are non-contact, effective in the dark, and acquire a direct measurement of signing kinematic via exploitation of the micro-Doppler effect. First, this work provides an in depth, comparative examination of the kinematic properties of signing as measured by RF sensors for both fluent ASL users and hearing imitation signers. Second, as ASL recognition techniques utilizing deep learning requires a large amount of training data, this work examines the effect of signing kinematics and subject fluency on adversarial learning techniques for data synthesis. Two different approaches for the synthetic training data generation are proposed: 1) adversarial domain adaptation to minimize the differences between imitation signing and fluent signing data, and 2) kinematically-constrained generative adversarial networks for accurate synthesis of RF signing signatures. The results show that the kinematic discrepancies between imitation signing and fluent signing are so significant that training on data directly synthesized from fluent RF signers offers greater performance (93% top-5 accuracy) than that produced by adaptation of imitation signing (88% top-5 accuracy) when classifying 100 ASL signs.

An Efficient Hierarchical Optic Disc and Cup Segmentation Network Combined with Multi-task Learning and Adversarial Learning.

Automatic and accurate segmentation of optic disc (OD) and optic cup (OC) in fundus images is a fundamental task in computer-aided ocular pathologies diagnosis. The complex structures, such as blood vessels and macular region, and the existence of lesions in fundus images bring great challenges to the segmentation task. Recently, the convolutional neural network-based methods have exhibited its potential in fundus image analysis. In this paper, we propose a cascaded two-stage network architecture for robust and accurate OD and OC segmentation in fundus images. In the first stage, the U-Net like framework with an improved attention mechanism and focal loss is proposed to detect accurate and reliable OD location from the full-scale resolution fundus images. Based on the outputs of the first stage, a refined segmentation network in the second stage that integrates multi-task framework and adversarial learning is further designed for OD and OC segmentation separately. The multi-task framework is conducted to predict the OD and OC masks by simultaneously estimating contours and distance maps as auxiliary tasks, which can guarantee the smoothness and shape of object in segmentation predictions. The adversarial learning technique is introduced to encourage the segmentation network to produce an output that is consistent with the true labels in space and shape distribution. We evaluate the performance of our method using two public retinal fundus image datasets (RIM-ONE-r3 and REFUGE). Extensive ablation studies and comparison experiments with existing methods demonstrate that our approach can produce competitive performance compared with state-of-the-art methods.

Adversarial learning techniques for security and privacy preservation: A comprehensive review

AbstractIn recent years, the use of smart devices has increased exponentially, resulting in massive amounts of data. To handle this data, effective data storage and management has required. Cloud computing (CC) is a promising solution to deal with this huge amount of data. Electronic devices are collecting real‐time data from sensors and applications through a wireless communication channel in the digital era. In some cases, CC cannot protect against various malicious attacks in the wireless communication channel. To address this issue, we have used machine learning (ML) and deep learning (DL) techniques for attack detection in a wireless channel on an early basis. It trains a model to predict malicious activities of attackers, which aids in the security of CC's sensitive data. We employed adversarial learning techniques (AL) to add fake data into the model to ensure that the trained model was correct. The trained model can distinguish between the fake and real data from the training samples and improve the training samples' performance. AL provides different defense mechanisms to preserve the privacy of ML‐ and DL‐based model but does not ensure the system's robustness. To improve the system's robustness, we have used federated learning with blockchain technology to make a system more robust, reliable, accurate, and transparent. This integration aids in providing high‐graded security against adversarial attacks. This paper presents a comprehensive review to highlight the recent improvements in AL techniques. Moreover, we explored the various AL applications in security and privacy preservation. Finally, open research issues and future directions are discussed to show future research avenues.

Data Discretization and Decision Boundary Data Point Analysis for Unknown Attack Detection

Researchers have continuously sought effective ways to immediately detect unknown (zero-day) cyberattacks. Most current methods rely on pattern-recognition to identify known threats when they appear; however, some newer capabilities use machine-learning (ML) anomaly detection tools that involve training a model based on normal network data, so the outliers can be identified and scrutinized in case they represent a new attack. Various deep-learning methods have been attempted for the latter, but training for unknown features and new events is problematic with machine learning (ML), owing to the need to train for the unknown, which actually increases the risk of false positives. Moreover, attacks developed using adversarial learning techniques can quickly outsmart such algorithms by hiding the attack in the normal distribution. To overcome these problems, this study applies data discretization and decision-boundary point analyses to scrutinize patterns near the thresholds of uncertainty. A novel discretization method is used to effectively train our model for the fuzzy c-means feature analysis of data points at the decision boundary, through which adversarial features are detected and classified based upon their entropy. Through this, it was possible to identify incorrectly detected attack data distributed near the model’s decision boundary. The National Security Laboratory’s Knowledge Discovery Dataset, which is commonly used to evaluate ML intrusion detection systems, is used to evaluate the proposed method. The results show that our model successfully identifies attacks at the decision boundary as desired and that its performance can be improved through classification. In addition, when classification was performed, it was confirmed that the accuracy performance of DoS attacks was improved by 5 to 7%, Probe by 7 to 10%, R2L by 4 to 7%, and U2R by 1 to 9%, compared to the existing model.

Fair and Effective Policing for Neighborhood Safety: Understanding and Overcoming Selection Biases.

An accurate crime prediction and risk estimation can help improve the efficiency and effectiveness of policing activities. However, reports have revealed that biases like racial prejudice could exist in policing enforcement, and trained predictors may inherit them. In this work, we study the possible reasons and countermeasures to this problem, using records from the New York frisk and search program (NYCSF) as the dataset. Concretely, we provide analysis on the possible origin of this phenomenon from the perspective of risk discrepancy, and study it with the scope of selection bias. Motivated by theories in causal inference, we propose a re-weighting approach based on propensity score to balance the data distribution, with respect to the identified treatment: search action. Naively applying existing re-weighting approaches in causal inference is not suitable as the weight is passively estimated from observational data. Inspired by adversarial learning techniques, we formulate the predictor training and re-weighting as a min-max game, so that the re-weighting scale can be automatically learned. Specifically, the proposed approach aims to train a model that: 1) able to balance the data distribution in the searched and un-searched groups; 2) remain discriminative between treatment interventions. Extensive evaluations on real-world dataset are conducted, and results validate the effectiveness of the proposed framework.

Deep adversarial domain adaptation for breast cancer screening from mammograms

The early detection of breast cancer greatly increases the chances that the right decision for a successful treatment plan will be made. Deep learning approaches are used in breast cancer screening and have achieved promising results when a large-scale labeled dataset is available for training. However, they may suffer from a dramatic decrease in performance when annotated data are limited. In this paper, we propose a method called deep adversarial domain adaptation (DADA) to improve the performance of breast cancer screening using mammography. Specifically, our aim is to extract the knowledge from a public dataset (source domain) and transfer the learned knowledge to improve the detection performance on the target dataset (target domain). Because of the different distributions of the source and target domains, the proposed method adopts an adversarial learning technique to perform domain adaptation using the two domains. Specifically, the adversarial procedure is trained by taking advantage of the disagreement of two classifiers. To evaluate the proposed method, the public well-labeled image-level dataset Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) is employed as the source domain. Mammography samples from the West China Hospital were collected to construct our target domain dataset, and the samples are annotated at case-level based on the corresponding pathological reports. The experimental results demonstrate the effectiveness of the proposed method compared with several other state-of-the-art automatic breast cancer screening approaches.

Improving Domain-Adaptive Person Re-Identification by Dual-Alignment Learning With Camera-Aware Image Generation

Domain adaptation in person re-identification (re-ID) has always been challenging, especially for the lack of supervision information on the target domain. Existing methods generally introduced extra supervision by adversarial learning techniques, then added all the augmented data in the training process to optimize the re-ID model. However, the direct utilization of all the generated data not only increases additional computational cost but also ignores the potential correlation between the origin and generated data. In this article, we propose a novel dual-alignment learning framework (DAL) with camera-aware image generation to efficiently and effectively tackle this issue. Specifically, we propose a camera transfer matching module to generate additional training images with different camera styles, and construct the matching pairs with each containing a origin image and one corresponding camera transferred image. To strengthen the correlation of images for each matching pair, we align the pseudo-labels via clustering algorithm to reduce the pseudo-labels distribution discrepancy between the origin and generated images. Besides, to avoid model degeneration affected by some inaccurate pseudo-labels on unlabelled data, we maximize the mutual information to align the image feature representations of matching pair. The DAL allows us to decrease the camera variance and enhance the discrimination ability of re-ID model. Extensive experiments on three large-scale benchmarks demonstrate the superiority of DAL over state-of-the-art methods.

Adversarial Learning-Based Semantic Correlation Representation for Cross-Modal Retrieval

Cross-modal retrieval has become a hot issue in past years. Many existing works pay attentions on correlation learning to generate a common subspace for cross-modal correlation measurement, and others use adversarial learning technique to abate the heterogeneity of multimodal data. However, very few works combine correlation learning and adversarial learning to bridge the intermodal semantic gap and diminish cross-modal heterogeneity. This article proposes a novel cross-modal retrieval method, named Adversarial Learning based Semantic COrrelation Representation (ALSCOR), which is an end-to-end framework to integrate cross-modal representation learning, correlation learning, and adversarial. Canonical correlation analysis model, combined with VisNet and TxtNet, is proposed to capture cross-modal nonlinear correlation. Besides, intramodal classifier and modality classifier are used to learn intramodal discrimination and minimize the intermodal heterogeneity. Comprehensive experiments are conducted on three benchmark datasets. The results demonstrate that the proposed ALSCOR has better performance than the state of the arts.

Towards blending Physics-Based numerical simulations and seismic databases using Generative Adversarial Network

This paper presents a new strategy to blend the outcome of physics-based numerical simulations with massive but poorly-labelled experimental databases such as in-situ data routinely recorded for monitoring purposes. The proposed approach relies on a set of adversarial learning techniques with a twofold purpose: (1) finding two reduced-dimensional non-linear representations of both synthetic and experimental data; (2) training a stochastic generator of fake experimental responses conditioned by the physics-based simulation results.This methodology is applied to earthquake ground motion prediction. Indeed, regional three-dimensional high-fidelity numerical models accounting for both extended sources and complex geology are still limited to a low-frequency range. Moreover, they are prone to significant uncertainties induced by a lack of data on small scale geological structures and rupture processes. Databases of broadband seismic signals recorded worldwide at seismological networks are used to retrieve some pieces of information on these small scale data to generate realistic broadband signals from synthetic ones.Outstanding performances in encoding seismic signals are demonstrated, together with efficient generation capabilities, provided that the physics-based results carry enough information to properly condition the stochastic generator. In addition, this paper shows that the proposed method, fed only with raw data from both databases and numerical models, outperforms other random signal generators based on pre-existing expertise such as prescribed spectra and more or less complex phenomenological models.

Learning adaptive geometry for unsupervised domain adaptation

Unsupervised domain adaptation is an effective approach to solve the problem of dataset bias. However, most existing unsupervised domain adaptation methods assume that the geometry structures of data distributions are similar in the source and target domains. This assumption is invalid in many practical applications, because the training and test datasets usually differ in the variability modes and/or variation degrees. This paper handles the problem of inconsistent geometries by aligning both data representations and geometries. To overcome the lack of target labels in aligning geometries, this paper proposes learning the adaptive geometry that is derived from the domain-shared label space. Source and target geometries are aligned by constraining them with the unified criteria of the adaptive geometry. Combining the adaptive geometry learning and adversarial learning techniques, we develop a geometry-aware dual-stream network to learn the geometry-aligned representations. Experimental results show that our method achieves good performance on cross-dataset recognition tasks.

Treant: training evasion-aware decision trees

Despite its success and popularity, machine learning is now recognized as vulnerable to evasion attacks, i.e., carefully crafted perturbations of test inputs designed to force prediction errors. In this paper we focus on evasion attacks against decision tree ensembles, which are among the most successful predictive models for dealing with non-perceptual problems. Even though they are powerful and interpretable, decision tree ensembles have received only limited attention by the security and machine learning communities so far, leading to a sub-optimal state of the art for adversarial learning techniques. We thus propose Treant, a novel decision tree learning algorithm that, on the basis of a formal threat model, minimizes an evasion-aware loss function at each step of the tree construction. Treant is based on two key technical ingredients: robust splitting and attack invariance, which jointly guarantee the soundness of the learning process. Experimental results on publicly available datasets show that Treant is able to generate decision tree ensembles that are at the same time accurate and nearly insensitive to evasion attacks, outperforming state-of-the-art adversarial learning techniques.