Publications

2026

1. ICMR

   MOSAIV: Multi-Agent LLM Swarms for Automated Multimedia News Verification: Fake News Detection

   Muhammad Shahid Muneer, Khoa Tran, Van Tuan Nguyen, and Simon S. Woo

   In Proceedings of the 2026 International Conference on Multimedia Retrieval, 2026

   Abs DOI

   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.

2. CVPR

   Efficient Unlearning through Maximizing Relearning Convergence Delay

   Khoa Tran, and Simon S. Woo

   In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Findings, Jun 2026

   Abs

   Machine unlearning poses challenges in removing mislabeled, contaminated, or problematic data from a pretrained model. Current unlearning approaches and evaluation metrics are solely focused on model predictions, which limits insight into the model’s true underlying data characteristics. To address this issue, we introduce a new metric called relearning convergence delay, which captures both changes in weight space and prediction space, providing a more comprehensive assessment of the model’s understanding of the forgotten dataset. This metric can be used to assess the risk of forgotten data being recovered from the unlearned model. Based on this, we propose the Influence Eliminating Unlearning framework, which removes the influence of the forgetting set by degrading its performance and incorporates weight decay and injecting noise into the model’s weights, while maintaining accuracy on the retaining set. Extensive experiments show that our method outperforms existing metrics and our proposed relearning convergence delay metric, approaching ideal unlearning performance. We provide theoretical guarantees, including exponential convergence and upper bounds, as well as empirical evidence of strong retention and resistance to relearning in both classification and generative unlearning tasks.

2025

1. WWW

   Fairness and Robustness in Machine Unlearning

   Khoa Tran, and Simon S. Woo

   In Companion Proceedings of the ACM on Web Conference 2025, Sydney NSW, Australia, Jun 2025

   Abs DOI

   Machine unlearning poses the challenge of ”how to eliminate the influence of specific data from a pretrained model” in regard to privacy concerns. While prior research on approximated unlearning has demonstrated accuracy and efficiency in time complexity, we claim that it falls short of achieving exact unlearning, and we are the first to focus on fairness and robustness in machine unlearning algorithms. Our study presents fairness Conjectures for a well-trained model, based on the variance-bias trade-off characteristic, and considers their relevance to robustness. Our Conjectures are supported by experiments conducted on the two most widely used model architectures-ResNet and ViT-demonstrating the correlation between fairness and robustness: the higher fairness-gap is, the more the model is sensitive and vulnerable. In addition, our experiments demonstrate the vulnerability of current state-of-the-art approximated unlearning algorithms to adversarial attacks, where their unlearned models suffer a significant drop in accuracy compared to the exact-unlearned models. We claim that our fairness-gap measurement and robustness metric should be used to evaluate the unlearning algorithm. Furthermore, we demonstrate that unlearning in the intermediate and last layers is sufficient and cost-effective for time and memory complexity.

2. Journal

   Gaussian mixture model and Bayesian convolutional neural network for abnormal musculoskeletal radiographs classification

   Xiem HoangVan, and Khoa Tran

   Evolutionary Intelligence, Jun 2025

   Abs PDF

   Bayesian convolutional neural network (BCNN) has recently demonstrated its efficiency in medical image classification. In its structure, BCNN formulates the value of neural network (NN) weights into a Gaussian distribution and creates the uncertainty estimation for the model. However, with abnormal classification in musculoskeletal radiographs, we unveil that only a Gaussian distribution is not enough to capture the weight characteristics. Therefore, BCNN model may not be suitable for abnormal musculoskeletal radiographs classification. In this context, we propose a novel Gaussian mixture model-based BCNN (MixBCNN) for abnormal musculoskeletal radiographs classification problem. In the proposed method, the distribution of NN weights is represented as a Gaussian mixture distribution, expecting a new neural network model with better representation. In addition, we propose an ensemble learning approach to efficiently combine potential NN backbones, notably the Densenet121, Densenet169 and Densenet201. Performance evaluation is examined for the popular MURA dataset and shows the superiority of the proposed model with respect to stateof-the-art BCNN methods, notably by 0.86 in F1 score and 0.72 in Cohens kappa metrics while asking for negligible complexity overhead.

2021

1. SMA

   Joint Image Denoising and Colorization Using Deep Network

   Khoa Tran, Dinh Quang Vinh, Nguyen Hong Phuc, Debnath C Narayan, NGUYEN Tuan-Duc, and Chang Wook Ahn

   In The 9th International Conference on Smart Media and Applications, Jeju, Republic of Korea, Jun 2021

   Abs DOI

   This paper significantly enhances from the work [1] and proposes a deep neural network that solves the denoising and colorization problem simultaneously. The joint problem is solved by two separate sub-networks that are trained in an end-to-end manner. Specifically, map attention modules are used to revise feature maps, while a few convolutional layers to extract features at the beginning of the network helps to boost the proposed network significantly. We use KITTI dataset to prepare training and testing datasets. In addition, we compare the proposed method with the baseline method using the PSNR and SSIM metrics. To have a fair comparison, we train the proposed and baseline methods using the same dataset, loss function, and training configurations. The experimental results show that the proposed method performed significantly better the baseline method in the KITTI dataset.