Verifying social media content from active conflict zones requires rapid geolocation, source attribution, forensic analysis, and multi-platform verification—tasks that overwhelm individual analysts at scale. We present Multi-agent OSINT Swarm for Automated Information Verification (MOSAIV), a three-stage agentic swarm built on Large Language Models (LLMs) for automated multimedia news verification. MOSAIV operates in three sequential phases: (1) a Prime Agent use 50 labeled training samples via few-shot in-context learning to produce a shared context document and a reusable 7-step verification skill specification; (2) multiple parallel Verification Agents each process social media posts using the primed skill, performing live web searches, OSINT analysis, and structured report generation; and (3) a dedicated Localization Agent independently verifies GPS coordinates and produces bounding-box-annotated evidence images, dual-panel OpenStreetMap location cards, and live source evidence thumbnails, multiple evidence artifacts per run in total. We evaluated 10 conflict-zone validation cases provided by the MV2026 challenge. MOSAIV achieves 10/10 precise GPS coordinates, 100% forensic analysis coverage, and an average of 8.1 independent sources per report, demonstrating that few-shot priming and agent swarms are critical for high-quality automated OSINT verification. We further conduct the first systematic model-capability scaling study across six Claude variants, identifying GPS precision, forensic completeness, and propaganda context detection.

Machine unlearning seeks to remove the influence of selected training examples without retraining the model from scratch. Recent work has extended this goal to vision–language models, yet existing datasets are not suited for judging whether a sample's influence has truly been erased from learned image–text pairs. Current algorithms often introduce false information into sentences generated after unlearning, which compromises utility. We first establish three criteria that an image caption unlearning method should meet: Specificity Reduction, Identity Removal, and Performance Preservation. Guided by these criteria, we present CelebCaption, an image–text dataset of 15,000 photographs covering 150 well-known individuals, each linked to four captions that vary in detail (detailed versus summary) and in the presence of the subject's name. This design enables controlled, quantitative assessment of the proposed unlearning objectives. We benchmark several representative unlearning algorithms on CelebCaption using both caption quality scores and membership inference attack accuracy as quantitative unlearning metrics, and observe that current methods fail to achieve their privacy objectives. Our unlearning criteria and dataset provide a focused, reproducible testbed for advancing privacy-aware image captioning. The CelebCaption dataset is publicly available at https://github.com/DASH-Lab/CelebCaption.

To address the potential risks of AI while supporting innovation and ensuring responsible adoption, there is an urgent need for clear governance frameworks grounded in human-centric values. It is imperative that AI systems operate in ways that are transparent, trustworthy, and ethically sound. Developing truly human-centric AI goes beyond technical innovation. It requires interdisciplinary collaboration and diverse perspectives. This workshop will explore key challenges and emerging solutions in the development of human-centric AI, with a focus on explainability, trustworthiness, fairness, and privacy. We welcome both theoretical contributions and practical case studies that demonstrate how human-centered principles are realized in real-world AI systems. The official workshop webpage is available at https://xai.kaist.ac.kr/Workshop/hcai2025/, which provides comprehensive information about the program.

Machine unlearning has emerged as a significant research topic in response to the increasing demands for data privacy and compliance with privacy regulations. The main challenge is to eliminate the influence of a specific subset of training data from a pretrained model while preserving the model's performance on the retain set without retraining the model from scratch. In this paper, we propose a novel efficient unlearning framework based on Optimal Transport, which can effectively work on both class-wise and instance-wise unlearning tasks. By analyzing and comparing the feature spaces of the original and retrained models, we formulate the unlearning problem as a distribution alignment task between the forget set and the retain set. We guide the feature distribution of the forget set, which initially forms distinct and structured patterns, to align with that of the retain set. Extensive experiments on three public benchmark datasets demonstrate its superior effectiveness compared to previous state-of-the-art methods.

Multi-step or hybrid deepfakes, created by sequentially applying different deepfake creation methods such as Face-Swapping, GAN-based generation, and Diffusion methods, can pose an emerging and unforseen technical challenge for detection models trained on single-step forgeries. While prior studies have mainly focused on detecting isolated single manipulation, little is known about the detection model behavior under such compositional, hybrid, and complex manipulation pipelines. In this work, we introduce \textbf{FakeChain}, a large-scale benchmark comprising 1-, 2-, and 3-Step forgeries synthesized using five state-of-the-art representative generators. Using this approach, we analyze detection performance and spectral properties across hybrid manipulation at different step, along with varying generator combinations and quality settings. Surprisingly, our findings reveal that detection performance highly depends on the final manipulation type, with F1-score dropping by up to \textbf{58.83\%} when it differs from training distribution. This clearly demonstrates that detectors rely on last-stage artifacts rather than cumulative manipulation traces, limiting generalization. Such findings highlight the need for detection models to explicitly consider manipulation history and sequences. Our results highlight the importance of benchmarks such as FakeChain, reflecting growing synthesis complexity and diversity in real-world scenarios. Our sample code is available here\footnote{https://github.com/minjihh/FakeChain}.

Open-vocabulary object detection (OVD) in remote sensing (RS) has shown remarkable generalization capabilities across diverse RS imagery through alignment between image and text embeddings. Such methods have further improved detection performance by incorporating additional scene-level context from both visual and textual domains. However, existing methods approximate scene context by simply averaging the text embeddings of the image's object labels, which is insufficient to capture the rich linguistic context present in RS scenes. To address this limitation, we propose a novel Feature-Aligned Scene Encoding (FASE), which constructs comprehensive scene representations through high-quality captions generated by a specialized vision-language model. Our Feature Alignment Module (FAM) creates a robust scene representation by fusing domain-specific caption embeddings with general text features through dual-branch fusion with gating and cross-attention. This resulting representation then facilitates the alignment with visual features. By utilizing enhanced scene encoding only during training, our method internalizes rich contextual knowledge without increasing inference complexity. Experiments on multiple benchmarks demonstrate significant improvements over state-of-the-art methods, validating the effectiveness of our approach for OVD in RS.

The rapid advancement of generative AI has enabled the mass production of photorealistic synthetic images, blurring the boundary between authentic and fabricated visual content. This challenge is particularly evident in deepfake scenarios involving facial manipulation, but also extends to broader AI-generated content (AIGC) cases involving fully synthesized scenes. As such content becomes increasingly difficult to distinguish from reality, the integrity of visual media is under threat. To address this issue, we propose a physically interpretable deepfake detection framework and demonstrate that defocus blur can serve as an effective forensic signal. Defocus blur is a depth-dependent optical phenomenon that naturally occurs in camera-captured images due to lens focus and scene geometry. In contrast, synthetic images often lack realistic depth-of-field (DoF) characteristics. To capture these discrepancies, we construct a defocus blur map and use it as a discriminative feature for detecting manipulated content. Unlike RGB textures or frequency-domain signals, defocus blur arises universally from optical imaging principles and encodes physical scene structure. This makes it a robust and generalizable forensic cue. Our approach is supported by three in-depth feature analyses, and experimental results confirm that defocus blur provides a reliable and interpretable cue for identifying synthetic images. We aim for our defocus-based detection pipeline and interpretability tools to contribute meaningfully to ongoing research in media forensics. The implementation is publicly available at: https://github.com/irissun9602/Defocus-Deepfake-Detection

For electric vehicles, the Adaptive Cruise Control (ACC) in Advanced Driver Assistance Systems (ADAS) is designed to assist braking based on driving conditions and user patterns. However, the driving data collected during development are limited and lack diversity, leading to late or aggressive braking. Moreover, it is necessary to effectively identify anomalies in braking patterns, which is critical for self-driving autonomous vehicles. We propose Graph Neural Controlled Differential Equation Normalizing Flow (GDFlow), which leverages Normalizing Flow (NF) with Neural Controlled Differential Equations (NCDE) to learn the distribution of normal driving patterns. Our approach captures spatio-temporal information from sensor data and accurately models continuous changes in driving patterns. Additionally, we introduce a quantile-based maximum likelihood objective to improve the likelihood estimate of normal data at the margin of the distribution. We validate GDFlow using real-world electric vehicle driving data that we collected from Hyundai IONIQ5 and GV80EV. Our model achieves state-of-the-art (SOTA) performance compared to nine baselines across four dataset configurations of different vehicle types and drivers. Furthermore, our model outperforms the latest anomaly detection methods across four time series benchmark datasets. Our approach demonstrates superior efficiency in inference time compared to existing methods.

Audio-visual deepfake detection methods demonstrate strong performance on academic datasets but fail significantly when applied to real-world. To address the shortcomings of previous approaches, we utilize landmarks dynamic information. First, we propose Landmark-based Distillation (LBD), motivated by I-JEPA's representation learning approach. LBD utilizes KL-divergence to align facial landmark predictions from visual and audio encoders, enforcing focus on geometric facial features rather than spurious background information. Second, we introduce Multimodal Temporal Information Alignment (MTIA), which employs contrastive learning to enhance temporal consistency between audio and visual representations. We conduct experiments on academic datasets and web-based deepfakes collected from diverse social media platforms, serving as real-world examples. Our proposed landmark-guided distillation framework achieves computational efficiency while improving multimodal video deepfake detection performance across a diverse range of deepfakes compared to existing methods. The code is available at https://github.com/Ckck12/Beyond-Masking.

The rapid development and prevalence of generative AI have made it easy for people to create high-quality deepfake images and videos, but their abuses have also increased exponentially. To mitigate potential social disruption, it is crucial to quickly detect the authenticity of each deepfake content hidden in a sea of information. While researchers have worked on developing deep learning-based methods, the deepfake datasets utilized in these studies are far from the real world in terms of their qualities; most popular deepfake datasets are human-distinguishable. To address this problem, we present a novel deepfake dataset, HiDF, a high-quality and human-indistinguishable deepfake dataset consisting of 62K images and 8K videos. HiDF is a meticulously curated dataset that includes diverse subjects that have undergone rigorous quality checks. A comparison of the quality between HiDF and existing deepfake datasets demonstrates that HiDF is human-indistinguishable. Hence, it can be a valuable benchmark dataset for deepfake detection tasks. Data and code (https://github.com/DSAIL-SKKU/HiDF) are publicly available for future deepfake detection research.

In the wake of a fabricated explosion image at the Pentagon, an ability to discern real images from fake counterparts has never been more critical. Our study introduces a novel multi-modal approach to detect AI-generated images amidst the proliferation of new generation methods such as Diffusion models. Our method, UGAD, encompasses three key detection steps: First, we transform the RGB images into YCbCr channels and apply an Integral Radial Operation to emphasize salient radial features. Secondly, the Spatial Fourier Extraction operation is used for a spatial shift, utilizing a pre-trained deep learning network for optimal feature extraction. Finally, the deep neural network classification stage processes the data through dense layers using softmax for classification. Our approach significantly enhances the accuracy of differentiating between real and AI-generated images, as evidenced by a 12.64% increase in accuracy and 28.43% increase in AUC compared to existing state-of-the-art methods.

Photo restoration technology enables preserving visual memories in photographs. However, physical prints are vulnerable to various forms of deterioration, ranging from physical damage to loss of image quality, etc. While restoration by human experts can improve the quality of outcomes, it often comes at a high price in terms of cost and time for restoration. In this work, we present the AI-based photo restoration framework composed of multiple stages, where each stage is tailored to enhance and restore specific types of photo damage, accelerating and automating the photo restoration process. By integrating these techniques into a unified architecture, our framework aims to offer a one-stop solution for restoring old and deteriorated photographs. Furthermore, we present a novel old photo restoration dataset because we lack a publicly available dataset for our evaluation.

We present Blind-Match, a novel biometric identification system that leverages homomorphic encryption (HE) for efficient and privacy-preserving 1:N matching. Blind-Match introduces a HE-optimized cosine similarity computation method, where the key idea is to divide the feature vector into smaller parts for processing rather than computing the entire vector at once. By optimizing the number of these parts, Blind-Match minimizes execution time while ensuring data privacy through HE. Blind-Match achieves superior performance compared to state-of-the-art methods across various biometric datasets. On the LFW face dataset, Blind-Match attains a 99.63% Rank-1 accuracy with a 128-dimensional feature vector, demonstrating its robustness in face recognition tasks. For fingerprint identification, Blind-Match achieves a remarkable 99.55% Rank-1 accuracy on the PolyU dataset, even with a compact 16-dimensional feature vector, significantly outperforming the state-of-the-art method, Blind-Touch, which achieves only 59.17%. Furthermore, Blind-Match showcases practical efficiency in large-scale biometric identification scenarios, such as Naver Cloud's FaceSign, by processing 6,144 biometric samples in 0.74 seconds using a 128-dimensional feature vector.

As the focus of space exploration shifts from national agencies to private companies, the interest in space industry has been steadily increasing. With the increasing number of satellites, the risk of collisions between satellites and space debris has escalated, potentially leading to significant property and human losses. Therefore, accurately modeling the orbit is critical for satellite operations. In this work, we propose the Decomposed Attention Segment Recurrent Neural Network (DASR) model, adding two key components, Multi-Head Attention and Tensor Train Decomposition, to SegRNN for orbit prediction. The DASR model applies Multi-Head Attention before segmenting at input data and before the input of the GRU layers. In addition, Tensor Train (TT) Decomposition is applied to the weight matrices of the Multi-Head Attention in both the encoder and decoder. For evaluation, we use three real-world satellite datasets from the Korea Aerospace Research Institute (KARI), which are currently operating: KOMPSAT-3, KOMPSAT-3A, and KOMPSAT-5 satellites. Our proposed model demonstrates superior performance compared to other SOTA baseline models. We demonstrate that our approach has 94.13% higher predictive performance than the second-best model in the KOMPSAT-3 dataset, 89.79% higher in the KOMPSAT-3A dataset, and 76.71% higher in the KOMPSAT-5 dataset.

The emergence of hyper-realistic deepfake videos has raised significant concerns regarding their potential misuse. However, prior research on deepfake detection has primarily focused on image-based approaches, with little emphasis on video. With the advancement of generation techniques enabling intricate and dynamic manipulation of entire faces as well as specific facial components in a video sequence, capturing dynamic changes in both global and local facial features becomes crucial in detecting deepfake videos. This paper proposes a novel sequential attentive face embedding, SAFE, that can capture facial dynamics in a deepfake video. The proposed SAFE can effectively integrate global and local dynamics of facial features revealed in a video sequence using contrastive learning. Through a comprehensive comparison with the state-of-the-art methods on the DFDC (Deepfake Detection Challenge) dataset and the FaceForensic++ benchmark, we show that our model achieves the highest accuracy in detecting deepfake videos on both datasets.

Machine learning has evolved through extensive data usage, including personal and private information. Regulations like GDPR highlight the "Right to be forgotten" for user and data privacy. Research in machine unlearning aims to remove specific data from pre-trained models. We introduce a novel two-step unlearning method, UNDO. First, we selectively disrupt the decision boundary of forgetting data at the coarse-grained level. However, this can also inadvertently affect the decision boundary of other remaining data, lowering the overall performance of classification task. Hence, we subsequently repair and refining the decision boundary for each class at the fine-grained level by introducing a loss for maintain the overall performance, while completely removing the class. Our approach is validated through experiments on two datasets, outperforming other methods in effectiveness and efficiency.

Though digital financial systems have provided users with convenient and accessible services, such as supporting banking or payment services anywhere, it is necessary to have robust security to protect against identity misuse. Thus, online digital identity (ID) verification plays a crucial role in securing financial services on mobile platforms. One of the most widely employed techniques for digital ID verification is that mobile applications request users to take and upload a picture of their own ID cards. However, this approach has vulnerabilities where someone takes pictures of the ID cards belonging to another person displayed on a screen, or printed on paper to be verified as the ID card owner. To mitigate the risks associated with fraudulent ID card verification, we present a novel dataset for classifying cases where the ID card images that users upload to the verification system are genuine or digitally represented. Our dataset is replicas designed to resemble real ID cards, making it available while avoiding privacy issues. Through extensive experiments, we demonstrate that our dataset is effective for detecting digitally represented ID card images, not only in our replica dataset but also in the dataset consisting of real ID cards. Our dataset is available at https://github.com/DASH-Lab/idcard_fraud_detection.

In recent times, there has been a notable surge in the amount of vision and sensing/time-series data obtained from drones and satellites.This data can be utilized in various fields, such as precision agriculture, disaster management, environmental monitoring, and others.However, the analysis of such data poses significant challenges due to its complexity, heterogeneity, and scale.Furthermore, it is critical to identify anomalies and maintain/monitor the health of drones and satellite systems to enable the aforementioned applications and sciences.This workshop presents an excellent opportunity to explore solutions that specifically target the detection of anomalies and novel occurrences in drones and satellite systems and their data.For more information, visit our website at https://sites.google.com/view/ansd23.

YouTube videos are one of the most effective platforms for disseminating creative material and ideas, and they appeal to a diverse audience. Along with adults and older children, young children are avid consumers of YouTube materials. Children often lack means to evaluate if a given content is appropriate for their age, and parents have very limited options to enforce content restrictions on YouTube. Young children can thus become exposed to inappropriate content, such as violent, scary or disturbing videos on YouTube. Previous studies demonstrated that YouTube videos can be classified into appropriate or inappropriate for young viewers using video metadata, such as video thumbnails, title, comments, etc. Metadata-based approaches achieve high accuracy, but still have significant misclassifications, due to the reliability of input features. In this paper, we propose a fusion model, called Samba, which uses both metadata and video subtitles for content classification. Using subtitles in the model helps better infer the true nature of a video improving classification accuracy. On a large-scale, comprehensive dataset of 70K videos, we show that Samba achieves 95% accuracy, outperforming other state-of-the-art classifiers by at least 7%. We also publicly release our dataset.

Knowledge distillation (KD) is a powerful technique for improving the performance of a small model by leveraging the knowledge of a larger model. Despite its remarkable performance boost, KD has a drawback with the substantial computational cost of pre-training larger models in advance. Recently, a method called self-knowledge distillation has emerged to improve the model's performance without any supervision. In this paper, we present a novel plug-in approach called Sliding Cross Entropy (SCE) method, which can be combined with existing self-knowledge distillation to significantly improve the performance. Specifically, to minimize the difference between the output of the model and the soft target obtained by self-distillation, we split each softmax representation by a certain window size, and reduce the distance between sliced parts. Through this approach, the model evenly considers all the inter-class relationships of a soft target during optimization. The extensive experiments show that our approach is effective in various tasks, including classification, object detection, and semantic segmentation. We also demonstrate SCE consistently outperforms existing baseline methods.

Recently, dynamic pruning methods have been actively researched, as they have shown very effective and remarkable performance in reducing computation complexity of deep neural networks. Nevertheless, most dynamic pruning methods fail to achieve actual acceleration due to the extra overheads caused by indexing and weight-copying to implement the dynamic sparse patterns for every input sample. To address this issue, we propose Dynamic Pattern-based Pruning Network (DPPNet), which preserves the advantages of both static and dynamic networks. First, our method statically prunes the weight kernel into various sparse patterns. Then, the dynamic convolution kernel is generated via aggregating input-dependent attention weights and static kernels. Unlike previous dynamic pruning methods, our novel method dynamically fuses static kernel patterns, enhancing the kernel's representational power without additional overhead. Moreover, our dynamic sparse pattern enables an efficient process using BLAS libraries, accomplishing actual acceleration. We demonstrate the effectiveness of the proposed DPPNet on CIFAR and ImageNet, outperforming the state-of-the-art methods achieving better accuracy with lower computational cost. For example, on ImageNet classification, ResNet34 utilizing DPP module achieves state-of-the-art performance with 65.6% FLOPs reduction and the inference speed increased by 35.9% without loss in accuracy. Code is available at https://github.com/lee-gwang/DPPNet.

Although the collision of space objects not only incurs a high cost but also threatens human life, the risk of collision between satellites has increased, as the number of satellites has rapidly grown due to the significant interests in many space applications. However, it is not trivial to monitor the behavior of the satellite in real-time since the communication between the ground station and spacecraft is dynamic and sparse, and there is an increased latency due to the long distance. Accordingly, it is strongly required to predict the orbit of a satellite to prevent unexpected contingencies such as a collision. Therefore, the real-time monitoring and accurate orbit prediction are required. Furthermore, it is necessary to compress the prediction model, while achieving a high prediction performance in order to be deployable in the real systems. Although several machine learning and deep learning-based prediction approaches have been studied to address such issues, most of them have applied only basic machine learning models for orbit prediction without considering the size, running time, and complexity of the prediction model. In this research, we propose Selective Tensorized multi-layer LSTM (ST-LSTM) for orbit prediction, which not only improves the orbit prediction performance but also compresses the size of the model that can be applied in practical deployable scenarios. To evaluate our model, we use the real orbit dataset collected from the Korea Multi-Purpose Satellites (KOMPSAT-3 and KOMPSAT-3A) of the Korea Aerospace Research Institute (KARI) for 5 years. In addition, we compare our ST-LSTM to other machine learning-based regression models, LSTM, and basic tensorized LSTM models with regard to the prediction performance, model compression rate, and running time.

Time series anomaly detection is extensively studied in statistics, economics, and computer science. Over the years, numerous methods have been proposed for time series anomaly detection using deep learning-based methods. Many of these methods demonstrate state-of-the-art performance on benchmark datasets, giving the false impression that these systems are robust and deployable in many practical and industrial real-world scenarios. In this paper, we demonstrate that the performance of state-of-the-art anomaly detection methods is degraded substantially by adding only small adversarial perturbations to the sensor data. We use different scoring metrics such as prediction errors, anomaly, and classification scores over several public and private datasets ranging from aerospace applications, server machines, to cyber-physical systems in power plants. Under well-known adversarial attacks from Fast Gradient Sign Method (FGSM) and Projected Gradient Descent (PGD) methods, we demonstrate that state-of-the-art deep neural networks (DNNs) and graph neural networks (GNNs) methods, which claim to be robust against anomalies and have been possibly integrated in real-life systems, have their performance drop to as low as 0%. To the best of our understanding, we demonstrate, for the first time, the vulnerabilities of anomaly detection systems against adversarial attacks. The overarching goal of this research is to raise awareness towards the adversarial vulnerabilities of time series anomaly detectors.

Anomaly detection in high-dimensional time series is typically tackled using either reconstruction- or forecasting-based algorithms due to their abilities to learn compressed data representations and model temporal dependencies, respectively. However, most existing methods disregard the relationships between features, information that would be extremely useful when incorporated into a model. In this work, we introduce Fused Sparse Autoencoder and Graph Net (FuSAGNet), which jointly optimizes reconstruction and forecasting while explicitly modeling the relationships within multivariate time series. Our approach combines Sparse Autoencoder and Graph Neural Network, the latter of which predicts future time series behavior from sparse latent representations learned by the former as well as graph structures learned through recurrent feature embedding. Experimenting on three real-world cyber-physical system datasets, we empirically demonstrate that the proposed method enhances the overall anomaly detection performance, outperforming baseline approaches. Moreover, we show that mining sparse latent patterns from high-dimensional time series improves the robustness of the graph-based forecasting model. Lastly, we conduct visual analyses to investigate the interpretability of both recurrent feature embeddings and sparse latent representations.

Generating a deep learning-based fake video has become no longer rocket science. The advancement of automated Deepfake (DF) generation tools that mimic certain targets has rendered society vulnerable to fake news or misinformation propagation. In real-world scenarios, DF videos are compressed to low-quality (LQ) videos, taking up less storage space and facilitating dissemination through the web and social media. Such LQ DF videos are much more challenging to detect than high-quality (HQ) DF videos. To address this challenge, we rethink the design of standard deep learning-based DF detectors, specifically exploiting feature extraction to enhance the features of LQ images. We propose a novel LQ DF detection architecture, multi-scale Branch Zooming Network (BZNet), which adopts an unsupervised super-resolution (SR) technique and utilizes multi-scale images for training. We train our BZNet only using highly compressed LQ images and experiment under a realistic setting, where HQ training data are not readily accessible. Extensive experiments on the FaceForensics++ LQ and GAN-generated datasets demonstrate that our BZNet architecture improves the detection accuracy of existing CNN-based classifiers by 4.21% on average. Furthermore, we evaluate our method against a real-world Deepfake-in-the-Wild dataset collected from the internet, which contains 200 videos featuring 50 celebrities worldwide, outperforming the state-of-the-art methods by 4.13%.

Recent advancements in web-based multimedia technologies, such as face recognition web services powered by deep learning, have been significant. As a result, companies such as Microsoft, Amazon, and Naver provide highly accurate commercial face recognition web services for a variety of multimedia applications. Naturally, such technologies face persistent threats, as virtually anyone with access to deepfakes can quickly launch impersonation attacks. These attacks pose a serious threat to authentication services, which rely heavily on the performance of their underlying face recognition technologies. Despite its gravity, deepfake abuse involving commercial web services and their robustness have not been thoroughly measured and investigated. By conducting a case study on celebrity face recognition, we examine the robustness of black-box commercial face recognition web APIs and open-source tools against Deepfake Impersonation (DI) attacks. While the majority of APIs do not make specific claims of deepfake robustness, we find that authentication mechanisms may get built one top of them, nonetheless. We demonstrate the vulnerability of face recognition technologies to DI attacks, achieving respective success rates of 78.0% for targeted (TA) attacks; we also propose mitigation strategies, lowering respective attack success rates to as low as 1.26% for TA attacks with adversarial training.

Recently, an early exit network, which dynamically adjusts the model complexity during inference time, has achieved remarkable performance and neural network efficiency to be used for various applications. So far, many researchers have been focusing on reducing the redundancy of input sample or model architecture. However, they were unsuccessful at resolving the performance drop of early classifiers that make predictions with insufficient high-level feature information. Consequently, the performance degradation of early classifiers had a devastating effect on the entire network performance sharing the backbone. Thus, in this paper, we propose an Efficient Multi-Scale Feature Generation Adaptive Network (EMGNet), which not only reduced the redundancy of the architecture but also generates multi-scale features to improve the performance of the early exit network. Our approach renders multi-scale feature generation highly efficient through sharing weights in the center of the convolution kernel. Also, our gating network effectively learns to automatically determine the proper multi-scale feature ratio required for each convolution layer in different locations of the network. We demonstrate that our proposed model outperforms the state-of-the-art adaptive networks on CIFAR10, CIFAR100, and ImageNet datasets. The implementation code is available at https://github.com/lee-gwang/EMGNet

Deep learning-based video manipulation methods have become widely accessible to the masses. With little to no effort, people can quickly learn how to generate deepfake (DF) videos. While deep learning-based detection methods have been proposed to identify specific types of DFs, their performance suffers for other types of deepfake methods, including real-world deepfakes, on which they are not sufficiently trained. In other words, most of the proposed deep learning-based detection methods lack transferability and generalizability. Beyond detecting a single type of DF from benchmark deepfake datasets, we focus on developing a generalized approach to detect multiple types of DFs, including deepfakes from unknown generation methods such as DeepFake-in-the-Wild (DFW) videos. To better cope with unknown and unseen deepfakes, we introduce a Convolutional LSTM-based Residual Network (CLRNet), which adopts a unique model training strategy and explores spatial as well as the temporal information in a deepfakes. Through extensive experiments, we show that existing defense methods are not ready for real-world deployment. Whereas our defense method (CLRNet) achieves far better generalization when detecting various benchmark deepfake methods (97.57% on average). Furthermore, we evaluate our approach with a high-quality DeepFake-in-the-Wild dataset, collected from the Internet containing numerous videos and having more than 150,000 frames. Our CLRNet model demonstrated that it generalizes well against high-quality DFW videos by achieving 93.86% detection accuracy, outperforming existing state-of-the-art defense methods by a considerable margin.

Reducing false positives while detecting anomalies is of growing importance for various industrial applications and mission-critical infrastructures, including satellite systems. Undesired false positives can be costly for such systems, bringing the operation to a halt for human experts to determine if the anomalies are true anomalies that need to be mitigated. Although rule-based or machine learning-based anomaly detection approaches have been studied, a tensor-based decomposition method has not been extensively explored. In this work, we introduce an Integrative Tensor-based Anomaly Detection (ITAD) framework to detect anomalies in a satellite system with the goal of minimizing false positives. We construct 3rd-order tensors with telemetry data collected from the Korea Multi-Purpose Satellite-2 (KOMPSAT-2) and calculate the anomaly score using one of the component matrices obtained by applying CANDECOMP/PARAFAC decomposition to detect anomalies. Our result shows that our tensor-based approach outperforms existing methods, achieving higher accuracy and lower false positive rates. And we successfully deployed our anomaly detection system in real KOMPSAT-2 mission operation.

The recent paramount success of the gig economy has introduced new business opportunities in different areas such as food delivery service. However, there are food delivery ride abusers who break the company rule by driving unauthorized vehicles that are not stated in the contract. These abusers are particularly problematic because they break the transportation regulations and unfairly take more orders. However, detecting these abusers are challenging because of lack of labeled datasets and these anomalous abusers do not frequently occur compared to normal riders. Furthermore, sequential patterns of abusing behaviors are not easy to model.

Although pronounceability can improve password memorability, most existing password generation approaches have not properly integrated the pronounceability of passwords in their designs. In this work, we demonstrate several shortfalls of current pronounceable password generation approaches, and then propose, ProSemPass, a new method of generating passwords that are pronounceable and semantically meaningful. In our approach, users supply initial input words and our system improves the pronounceability and meaning of the user-provided words by automatically creating a portmanteau. To measure the strength of our approach, we use attacker models, where attackers have complete knowledge of our password generation algorithms. We measure strength in guess numbers and compare those with other existing password generation approaches. Using a large-scale IRB-approved user study with 1,563 Amazon MTurkers over 9 different conditions, our approach achieves a 30% higher recall than those from current pronounceable password approaches, and is stronger than the offline guessing attack limit.

A package tracking number (PTN) is widely used to monitor and track a shipment. Through the lenses of security and privacy, however, a package tracking number can possibly reveal certain personal information, leading to security and privacy breaches. In this work, we examine the privacy issues associated with online package tracking systems used in the top three most popular package delivery service providers (FedEx, DHL, and UPS) in the world and found that those websites inadvertently leak users’ personal data with a PTN. Moreover, we discovered that PTNs are highly structured and predictable. Therefore, customers’ personal data can be massively collected via PTN enumeration attacks. We analyzed more than one million package tracking records obtained from Fedex, DHL, and UPS, and showed that within 5 attempts, an attacker can efficiently guess more than 90% of PTNs for FedEx and DHL, and close to 50% of PTNs for UPS. In addition, we present two practical attack scenarios: 1) to infer business transactions information and 2) to uniquely identify recipients. Also, we found that more than 109 recipients can be uniquely identified with less than 10 comparisons by linking the PTN information with the online people search service, Whitepages.

Detecting fake or disguised face images become much more challenging due to the significant advancements made in machine learning, computer vision, and image processing techniques. In addition, due to the rise of various DeepFakes, fake images can be maliciously used to attack individuals and deter true information. Therefore, it is crucial to building a classifier that accurately distinguishes an individual from different or similar persons. In this preliminary work, we aim to detect a target person's face from different similar individuals, Doppelgangers, leveraging the dataset from Disguised Faces in the Wild (DFW) 2018. We use well-known off-the-shelf face detection classifiers, such as ShallowNet, VGG-16, and Xception to evaluate the classification performance. In order to further improve the detection performance, we apply data augmentation. Our preliminary result shows that the Xception model can classify one from different individuals with a 62% accuracy.

Manufacturing steel requires extremely challenging industrial processes. In particular, predicting the exact time instance of opening and closing tap-holes in a blast furnace has a great influence on steel production efficiency and operating cost, in addition to human safety. However, currently predicting the time to open and close tap-holes of the blast furnace still highly relies on manual human expertise and labor. Also, most of the prior research is limited to indirectly model the level of liquids in the hearth, using complex mathematical models or classical machine learning approaches.

Detecting an anomaly is not only important for many terrestrial applications on Earth but also for space applications. Especially, satellite missions are highly risky because unexpected hardware and software failures can occur due to sudden or unforeseen space environment changes. Anomaly detection and spacecraft health monitoring systems have heavily relied on human expertise to investigate whether they are a true anomaly or not. Also, it is practically infeasible to produce labels on data due to the enormous amount of telemetries generated from a satellite. In this work, we propose a data-driven anomaly detection algorithm for Korea Multi-Purpose Satellite 2 (KOMPSAT-2). We develop a Multivariate Convolution LSTM with Mixtures of Probabilistic Principal Component Analyzers, where our approach uses both neural networks and probabilistic clustering to improve the anomaly detection performance. We evaluated our approach with a total of 22 million telemetry samples collected for 10 months from KOMPSAT-2. We also compare our approach with other state-of-the-art approaches. We show that our proposed approach is 35.8% better in precision, and 18.2% better in F-1 score than the best baseline approach. We plan to deploy our algorithm in the second half of 2019 to actually apply real operation of KOMPSAT-2.

In the past, there have been several studies in analyzing password strength and structures. However, there are still many unknown questions to understand what really makes passwords both memorable and strong. In this work, we aim to answer some of these questions by analyzing password dataset through the lenses of data science and machine learning perspectives. We use memorable 3,260 password dataset collected from prior IRB-approved user studies over 3 years and classify passwords into three strength groups using online and offline attack limits. Then, we apply a tensor decomposition to analyze password dataset by constructing a 3rd-order tensor with passwords' syntactic and semantic features. In particular, we used PARAFAC2 tensor decomposition to uncover the main characteristics and features that affect password strength. We quantitatively identified the underlying factors that are more frequently observed in strong and memorable passwords. We hope that our finding can validate widely accepted advice for creating strong passwords and provide useful insights to design a better password suggestion system.